The PF diatonic algorithm has never been truly polyphonic, especially for shifting an interval like a diatonic 3rd. The only way to do this would be to have real time source separation of the guitar notes, and run two separate pitch shifters on each note (as far as I know, there is not a single product on the market that can do this yet, closest I've seen is Melodyne which isn't real time, and research stuff out of Universities). You could also use a hex pickup with separate pitchshifters.
I'll break it down a bit:
In F-major, the diatonic 3rd of an F note is the major 3rd interval, but the diatonic 3rd of the D note is a minor 3rd interval, so the pitchshifter has to make a decision, major or minor? It often goes with the signal that is stronger in the additive mix of the two notes, with the result that is out of key for the other note.
I promise that Version 2.3 would do the exact same thing, as what was broken in v3, and fixed in v3.5, was bug that messed up diatonic shifting at low levels. Now if you just want to do a straight major 3rd or minor 3rd (or any other interval), you should try PitchFlex or Harmodulator, as their lack of diatonic decision making lends them to be easier on polyphony.
What you're asking for is still cutting edge research that I and many, many other engineers have been chasing for years, so it's not an issue of fixing anything in the PF, rather, it's about doing something innovative. I truly hope we can deliver this someday in a product.