sfx2000
Part of the Furniture
even at low .ac sig str I still have ".ac" reported in the 3 brands of APs I'm currently testing.... it did not "fall back" to N.
If it were to fallback from AC to N, it would likely be a disconnect/re-associate - generally though, and I've only casually examined this - N450 vs. AC867 in 5GHz, (three stream 11n vs. 2 streams on 11ac), AC generally will provide a better link at the same distance and RSSI - at the extreme limit, it's more the adapter, it might hang on a bit longer...
Devices - MacBook Air 2014 (11ac 2-stream) vs. MacBook Pro 2012 non-Retina (3-stream N) - AP test was Airport Extreme AC.
What I can say, once the clients drop to 2.4GHz, the 3-stream client was significantly better off - but this again, it client dependent and RF dependent - the MBPro has a very good 2.4GHz radio (just right) and the MacBook Air has a higher gain 2-stream radio, but it gets noise limited due to the higher gain antennas in it compared to the MBP...
Again, nothing formal, just moving about the house, but interesting - if I were to get these on a bench - I would use a combo of Attenuation (RSSI) and an AGWN noise source (CINR impact from adjacent AP's) to map the crossover points and test from there. Getting deep, one could also drop in a raypath fader to look at Rician and Rayliegh fading impacts (position dependency at a give RSSI/CINR level).