Mechanism behind distortion at max output

When you look at the distortion vs power output of most class-D amplifiers, they typically are pretty flat until max power output is approached - then they head for the moon (10%).

What's the mechanism behind this? Voltage drop across output devices as they conduct larger currents?

I cant believe the modulator would be any less linear in the last few percent of duty cycle, versus the first - or a few percent around 50% (in the middle...)

Not that it's a problem I run into often; just curious how that works out from a circuit operation standpoint. Thanks ahead for any ideas on this!
 
My impression is that it relates to the difficulty of generating very low or very high duty-cycles that makes the THD-curve go up. Further attempt of forcing more power out should leave the modulation stuck at 0% or 100% at the peaks, similar to voltage saturation for class AB amplifiers.