Thank you Mr.Clark KISS is how I like things normally less smoke, and mirrors.
The 1N400X diodes could be a lower Vf but why worry, as you say solves the Vgs problem and now we can use a bigger FET, with a lower RdsOn but why the track has a higher resistance than the FET. Also changing the constant current Darling to a FET to give us another -2V swing and then we can change the sense and source resistors to lets say 10mR not forgeting Ohm's law as the current IQ will have to be reset to read 10mV/A not 100mV/A thus saving another 500 - 750mV now we could get up to a +20v to -20v swing, ie 14VRms about 24W out but the IQ would need to be set to about 3.5A and that now 140W of heat we need to dump.
So at 2.5A and 14V the nice load look like 10Ohm at about 16W but with much higher peaks. So the only limiting factor is getting rid of the heat, or a rethink on the PSU but then the heat will move to the regulators.
Now with only 100W to dump as heat. A slow fan maybe?
Now Alan Brown built one with huge IGBT,s 300A 1200V and monster heatsinks and we got close to 30W nice work Mr.Brown. With if I recall two 100,000uf cap on each rail and Mono Blocks.
He has a few circuit boards left over so chat him up.