1) I conducted extensive listening tests (single blind) at various configurations and differing amounts of crossover distortion were consistently audible - and there was consistently measurable and observable differences on a 'scope.
2) The listening tests were using music of course...
3) One obvious way to "measure" with any signal source is subtraction (nulling).
1) What level was distortion 'consistently audible'?
2) In reported tests (the earliest I know of off hand was mid '70s in
Stereo Review--Bob Carver modified one of his Phase Linear amps to inject increasing levels of distortion, and a panel listened in order to decide when it became audible. Audibility was initially heard on test tones, and only at higher levels with actual music. The more complex the music the less audible the distortion. I think thresholds were a couple of three percent on music. Might have been four percent on complex stuff.
3) Again, Carver 'demonstrated' this technique to both Peter Aczel and the gang at
Stereophile. Peter told me that his comparison (using the Mark Levinson ML-2 class A amp) was probably invalid anyway from an audibility standpoint, because prior to transfer function matching via nulling, the amps were likely indistinguishable if operated within their respective electrical parameters. At the time Aczel was not using matched levels to listen for differences. The
Stereophile drama could have been a different thing in the audibility department inasmuch as Bob was nulling against a tube amp. So in that test there was a big difference in the amps from the get-go, due to design topology and intrinsic electrical performance.
Side note: John Saiu's 'white paper' on the amplifier/loudspeaker interface (amplifier damping factor-cable-speaker) is now on-line. John provides a handy Excel formula that one can use to more or less decide about their setup. He acknowledges that FR changes in or around 0.1 plus or minus dB are probably not going to matter. The question of course is whether one's amp can handle the sometimes difficult load variance asked of it.