Well, I was born after a 41-week gestation and... Wait. Oh; sorry. From the title of the question, I mistakenly thought you were talking about me. <cough, cough>
The two women are almost certainly not false positives. I say "almost" only because I gather the info is only on AncestryDNA, where we can't see anything but the total amount shared and the number of shared segments. If we could get the detail on the matches--from an upload to FTDNA or GEDmatch, for example--we could confirm it absolutely. Generally speaking, I've never seen a segment over 30cM that didn't turn out to be half-identical by descent and that would survive trio phasing (actually over 25cM, but I'll hedge my bets).
At 317cM and 388cM on Ancestry, at first glance I would say that she's probably more correct than not: that's an unusually large amount of shared DNA for 3rd cousins. It's even well over the Shared cM Project's max-range for 3rds, and that's crowd-sourced data that includes all kinds of family trees, pedigree collapsed and not. Looking at the numbers only, I put the probabilities at 2nd cousin, 1st cousin 2x removed, or the less likely (generationally) 2g-grand-aunt/uncle or 2g-grand-niece/nephew.
You didn't mention how Ancestry reports the two ladies' matches to your father and your great-aunt. The percentage of consanguinity between any two individuals decreases fourfold as the most recent common ancestor recedes one generation, so if they match you as nominal 2nd cousins (expected average sharing 212.5cM), and if they're truly on your direct Ferraiolo line, then they would be expected to share about 850cM with your father if 1st cousins, or about 425cM if 1C1R.
My only recommendation would be to try to see if they would be willing to put their data at a service where you can view the chromosome and segment details. Then, if you haven't already, begin compiling your own chromosome maps--spreadsheet or DNA Painter or Genome Mate Pro, whatever you like best--so that you can ferret out which segments seem to be reaching you (and your father; having his results is a huge benefit) from whom and when. If there's pedigree collapse in play somewhere, the total amount of shared atDNA won't tell the whole picture. Contiguous segment length still tends to (mostly; generalizing here) break up with each birth event; if pedigree collapse isn't in immediate generations (i.e., parents or grandparents) the largest segment sizes may tell as much of the tale as the total amount shared. But you can't get that detail from AncestryDNA.
Looks like an interesting mystery! Have fun!