There also wasn't very much of a measurable difference and the
differences weren't all positive for class A. For example, running
an output stage in class A can increase distortion, because the
output devices have to run at higher current levels, where the
output transistors might be far less linear.

Might be. Are they? A well-known designer, a friend of mine, is emphatic on
this point -- they aren't. They're more linear. Or rather, less non-linear.

As my distortion measuring gear progressed to residuals below
0.01%, there was always a mixed bag of measurable differences.
But they were arguably quite small, given that its often darn hard
to hear distortion below 0.1% or so.

Probably so, but what types of distortion?

I'm going to try to get Clay Barclay to write something about Crown's
null-testing system. He probably won't, but it's worth a try.