The 15 AncestryDNA cases I have vary from 606cM to 966cM; average 816cM and standard deviation of 120cM.
So while 1006cM is outside of what I've seen (by a little bit), it's (1006cM-816cM)/120cM = 1.58 standard deviations above the average. For a normal distribution, 90% of the values should be within 1.64 standard deviations.
In other words, it's within the bounds of what you'd expect to see 90% of the time - kind of high, but not all that unusual.
Also, notice that the average of your three is 785cM, really not very far from my average.
Intermarrying will certainly increase the numbers, but if you went back three generations and still didn't see a common ancestor it's not going to amount to anything very significant.
It could just be that the two of you simply happened to get a somewhat above-average fraction of your genes from the same grandparent.
The shared cM project chart is notorious for having overly broad ranges, BTW. It lumps several testing companies together (which is bad, because they all do things differently, and the results come out a bit different), he averages in samples reported to have endogamy (which is just incompetent), etc. For 1C on AncestryDNA, the 99% interval should be something in the low 500cMs to something in the low 1100cMs.
