It is almost impossible to do an apples to apples comparison here. The power ratings of the Purifi and GaN amps are different, so comparing absolute numbers of watts isn't valid. Efficiency at a given lower power isn't really valid either, as again, the full power capabilities are different and they are at different points on their efficiency curves.
It also is not useful to compare efficiency percentages at these high efficiencies. It makes more sense to compare losses. A change from 88% to 90% might be 2.3% in efficiency, but it is a 20% drop in losses. Perspective matters.
Comparing losses at full power is far from the full story. Most amps never run continuous full power. It makes much more sense for actual use conditions to compare amplifiers at some sensible level driven with some reasonable representation of actual use signal. One issue is that there are other losses than switching losses. Comparing at full power tends to skew the importance of switching losses versus total system losses.
Efficiency isn't really the game anyway. Heat dissipation in real use is the important number. Efficiency just looks good on publicity material.
If GaN lives up to its hype, we might see a halving in switching losses. Other losses will probably remain the same. So how much we win in reality is hard to guess from the numbers available. But halving of losses is the best case. It is going to be continually harder and harder to make real gains.
As noted earlier, the Axgin AX5689 provides a single chip mutli-channel PWM driver solution that takes feedback from after the output filter. There is a good chance a proper implementation using that chip (not the SMSL attempt) is going to stir the pot.