I would guess that either flipping PR2 or PR3 would give nearly the same effect (g = 0.9) and that flipping both makes it even more stable (smaller g). But what we really need is to see the cavity scan / HOM resonance plot to compare the cases.
The difference of 0.5% in mode-matching is not a strong motivation to make a choice, but sensitivity to accidental HOM resonance of either the carrier or f1 or f2 sidebands would be. Should also check for 2*f2 and 2*f1 resonances since our modulation depth may be as high as 0.3. Accidental 2f resonance may disturb the 3f error signals.
You would guess, and I would have guessed too, but the calculations tell the story. Flipping both does not increase the stability. The main issue is that flipping PR3 induces considerable astigmatism. This is why flipping PR3 alone does not make the cavity stable. I will do some simple calculations today that will demonstrate this effect.
But again, we should only change one thing at a time and understand that before moving on. Given that the calculations show that flipping only PR2 should alone have a positive affect, we should do just that first, and verify that we understand what's going on, before we move on to making more changes.
I will try to make some more arbcav runs as well, for just the flipped PR2.