The point is why tune feedback using signal that contains harmonics into the hundreds of kHz and higher when, at least in this case, I would suspect the bandwidth of the OPTs aren't remotely capable anything near that.
The "Too Long; Didn't read" answer:
The bandwidth of the amplifier, even a lesser pedigreed one like the one being discussed on this thread, does have bandwidth that exceeds its published "flat response" specs. That, with the combination of applying global feedback creates conditions in which the amplifier may oscillate at certain frequencies, under certain speaker loads. It is for this reason that it's important to test stability/frequency response of the amplifier at frequencies well below and above the "flat" response the amp is capable of producing.
On the other hand if the amplfier had no feedback whatsoever, it would not (generally) be possible for it to oscillate. So it is primarily because of feedback that these things need to be tested.
===
The very long (and boring?) answer:
If the stages encompassing the feedback loop include R/C or L/C couplings, then as the signal passes through those couplings, the signal becomes phase shifted. It's not a constant phase shift, rather it's more pronounced at certain frequencies, typically at the frequency extremes. It's important to note that ALL feedback signals are phase shifted at some level IF the loop includes R/C or L/C couplings.
In traditional global feedback push pull amplifiers, a portion of the (phase shifted) output signal is feed back to the input of the amplifier where it is subtracted from the input signal (hence the name "negative" feedback). The idea is that under normal operating conditions, the feedback signal is essentially an "error" signal--a measure of the "error" the amplifier is producing relative to its input signal. Thus feedback, when correctly applied, by and large corrects the amplifier's faults and tries to produce that perfect output signal, which is the input signal, just amplified.
All this is well and good, and works beautifully, except for one small problem. What if there is so much phase shift in the feedback signal (relative to the input signal) that the feedback signal is no longer an "error" correcting signal but an "error" producing signal? That occurs when the feedback signal is phase shifted by 180 degrees or more (relative to the input signal). When this occurs you have conditions for an oscillating amplifier. Even if the amplifier only has flat response between 20 Hz and 20 KHz, it's still possible for the amplifier to oscillate at 2x, 3x, 4x of the lowest OR highest frequency the amplifier is capable of producing with no loss.
To address the problem, you "tune" the amplifier to essentially force the closed loop gain of the amplifier to be less than unity at frequencies where the phase shift exceeds 180 degrees, thus forcing the amplifier to not oscillate at those frequencies.
So, in summary, this is why it's important to test the amplifier with a signal that contains lots of harmonics above the 20 Khz region to ensure that it doesn't oscillate at those frequencies. You also do essentially the same type of test at very low frequencies to ensure the amplifier doesn't oscillate at those low frequencies.
If you don't test for this, it's possible the amp will oscillate on it's own under normal operating conditions (but only if it has feedback applied). And you don't want your test bed environment to be your expensive speakers!
