I think I'm beginning to understand what Jam meant about the topology. I tried increasing the current mirror resistance to drop the Vce (and therefore the Early effect). However, once you drop the voltage you no longer have enough to swing the output devices near the rails.
I did manage to drop the THD by about 80% (!) by increasing the current mirror resistance to 330R : 100R
and bumping the rail voltage by 2V. (Hey, it's just a bit more heat to burn off, right?) I don't
think this had anything to do with the Early effect because the two changes play against each other and leave Vce only very slightly lower.
I was much less successful homogenising the current mirrors. The TO-126s just don't have enough gain. The KSA1220s are better than 1381s, but to equal the closed-loop gain of the hybrid TO-92/TO-126 mirrors I had to reduce the feedback, frittering away all the THD gains of the homogenous current mirrors.
I think the next step is to try out the MOSFET second stage from Nelson's Sony VFET part 2 article. I'm sure the Sony schematic Jam posted has merit, but I struggle to even make heads or tails of it.
Cheers,
Jeff.