This is a tangent--sort of--so I thought about making it a comment rather than an answer. But it doesn't seem to to fit as a comment and, surprise!, it grew too long to allow it to muddle another conversation.
Valerie offered the good advice of checking the Shared cM Project tool at DNA Painter. Fairly recently, Jonny Perl, the author of DNA Painter, made the individual relationship boxes in that tool clickable in order to present a pop-up window of the histogram for that particular connection. I think those are very important to look at; Blaine Bettinger includes them in the full PDF report of the Shared cM Project, but it's easy just to consider the low-to-high value ranges presented and assume that they're all similarly accurate. The probabilities as shown from calculations by Leah Larkin (The DNA Geek) help a lot, but the data still are what they are.
The fact is, we have no experimental, peer-reviewed research to help substantiate what are and are not valid centiMorgan ranges, or even percentage sharing ranges...which would be somewhat more precise because centiMorgans can be calculated differently depending on at least three independent factors, and because the values we see from testing companies can start out calculating the whole genome differently, from a high of 7494cM to a low of 6800cM (some studies have calculated the male genome to 5618cM). If we get new information any time soon separate from the Shared cM Project, my bet is that it will come from the Williams Lab at Cornell University in the form of extended computer simulations.
But for now, the Shared cM Project is the best source there is, and even Blaine is very clear that, by its very nature, crowd-sourced data can be inaccurate and that there is no way to vet what input is and isn't valid. The reliance is on volume: with enough submissions by knowledgeable people, mistakes and errors can be sublimated because they should be far fewer by proportion.
I'm not entirely sure that's true when it comes to self-reporting of DNA and genealogy, but Blaine does take a manual route to help eliminate the most egregious problems: he removes, from the analysis of each relationship, 0.5% of submissions from the high end and 0.5% from the low end. This arrives at an approximation of a 99% confidence interval. For this last update, he also supplied an estimated standard deviation for each of the 10 major groupings he uses (down to 4C1R; he doesn't estimate farther removed than that). To me, this was just as important as Jonny making the histograms immediately available on the tool at DNA Painter. It's really rough, back-of-the-napkin stuff, but I took those standard deviations and made a table (a PDF file viewable here) by backing into tighter approximate confidence intervals of 95% and 68%.
Other than the most pervasive errors in the data--people submitting information that is incorrect either because they misname the genealogical relationship (e.g., a 3C rather than a 1C2R) or believe they have identified a distant cousinship for which the DNA is invalid and can't be substantiated--the problem with crowd-sourced data is that we can't know what we don't know. Most testing companies will stop reporting, reasonably so, at segment lengths smaller than 6cM to 8cM, depending upon the company. With GEDmatch you can go smaller, but since they do no phasing or imputation even segments shown as large as 10cM can be false a significant percentage of the time. The result is that the lowest values will always be underrepresented in the Shared cM Project--at least in more distant relationships, and we can see this starting to happen when the histograms begin to show counts that are decidedly to the left in the graphs, as with 3C1R--and the averages skewed to a sharing amount that is artificially high.
Deb's 3C1R relations are showing as 132cM, 182cM, and 205cM. The projected 68% confidence interval shows a range of 16cM to 80cM for that relationship, with a theoretical average of 27cM. Looking at Blaine's histogram, 96.9% of all 3C1R reported matches were 125cM or less; 99.7% were 175cM or less. So, yes: the values are out of bounds for 3C1R.
Genealogy is really tough to mix with DNA because we come into it by default carrying a great big bag of confirmation bias. Over the years I've communicated with folks about DNA "evidence" in everything from low-resolution mtDNA tests to fairly outlandish items like autosomal DNA triangulations out to 10th cousins...things where the odds are about like being bitten by a shark while being struck by lightning. It's our family, so of course we tend to think that, since outliers do exist, that we must be the outlier. But that obviously can't be the case or there would be no outliers. If something looks out of the ordinary, it's probably out of the ordinary.
Last up is that exceptionally large autosomal DNA sharing is often written off as pedigree collapse somewhere back in the tree. That definitely can have an effect, especially in populations that have been endogamous in the past couple of centuries. But pedigree collapse is in all our trees, and a few instances of it, even in the genealogical timeframe, don't necessarily show up to any great extent among current-generation test-takers. The genius of biology is that occasional pedigree collapse doesn't have much lasting impact...otherwise the species would never have survived extreme population bottlenecks that have occurred in history.
Speaking of extremes, think Game of Thrones. The Lannisters' Cersei and Jaime were fraternal twins...who share the same amount of DNA as full siblings. Their children, had Cersei and Jaime been unrelated, would normally have shared about half their DNA. But Joffrey and the other kids (can't remember their names) were also nieces/nephews of their parents. So Joffrey et al. would also be 1st cousins because their parents were brother and sister. The average 1C sharing is 12.5%, which means instead of sharing 50% with each other, the kids would have shared about 62.5%.
Say two of Cersei's and Jaime's children had kids of their own but, this time (thank goodness), the other two parents weren't related to the Lannisters or to each other. So the kids--Cersei's and Jaime's grandkids--would be double 1st cousins rather than plain ol' 1st cousins. Their genetic relationship would look a little more like a half-sibling than a 1C, and they'd match on about 23.4% of their autosomal DNA. In the imperfect world of centiMorgan calculations, that would mean somewhere around 1,592cM to 1,756cM rather than the 850cM to 937cM we would expect for 1st cousins.
Continuing that pedigree, if these grandchildren of Cersei's and Jaime's grandkids then had children of their own--again with unrelated parents; the pedigree collapse stopped with Cersei and Jaime--then those children would be double 2nd cousins to each other. Genetically, that's about the same amount of sharing we'd expect from 1C1R: about 6.25% instead of the 3.125% expected of regular 2nd cousins. In centiMorgans, we're looking at roughly 425cM to 468cM rather than the 212cM to 234cM expected of 2nd cousins.
When we get to Cersei's and Jaime's 2g-grandchildren in this same unrelated-parents progression, we're down to the genetic difference between double 3rd cousins and regular 3rd cousins, and a distinction of about 106-117cM versus 53-59cM. Not insignificant, but we're already to a point where the amount of shared DNA can't readily distinguish between the pedigree collapse scenario and the one where none of the parents were genetically related.
They key is that the pedigree collapse didn't continue, that half the DNA in each birth came from an unrelated source.
Repeated pedigree collapse, as in endogamous populations, can be extremely difficult to evaluate...which is why for recent instances of endogamy, as with the Rapa Nui people of Easter Island, autosomal DNA is generally useful only to a couple of generations previous. But the effects of even extreme instances of pedigree collapse, as with the Lannisters, is diluted in the gene pool fairly quickly if the subsequent parents are otherwise unrelated.
Net message here is that overlarge DNA sharing results, proportionately speaking, shouldn't be quickly written off by the possibility that some set of distant ancestors were related. In Deb's case, the sharing amounts are about six times greater than would be expected for a 3C1R relationship, or around three times greater than the top 63% of respondents to the Shared cM Project; two times greater than 84% of the respondents. If the most recent common ancestors really are the 2g-grandparents then, using the Lannister example, even if they had been brother and sister what we'd expect to see down at the 3C1R level would be a DNA sharing of 0.7813% instead of 0.3906%, or a difference of about 25cM.
Edited: Crikey! A terrible, disconnected sentence structure, a poor choice of words, and a typo. I shouldn't be allowed near a keyboard today...
Edited Again: Some rightly pointed out that, in Game of Thrones, Jamie and Cersei Lannister were twins, and I had written that they were full siblings. Being male and female they were, of course, fraternal or dizygotic twins, not identical or monozygotic twins. The amount of DNA they shared would be no different than would any male/female set of full siblings.
Edited Again Again: Working on a new post made me realize I'd made a computational mistake when describing the descendants of Cersei and Jaime Lannister. Hm. I wonder if anyone has ever written a genetic genealogy summary of the entire Game of Thrones major families based upon the lineages as described in the books...
Not a complete workup of the families in GoT, but I did just find this article, from Stanford's The Tech Interactive. And Cersei's and Jaime's children were Joffrey, Tommen, and Mycella...whose names I will now promptly forget again.