Secondly, logically speaking, THD is not that important in an amp, less THD means better design of course, but speakers (or headphones) usually have a THD an order of magnitude superior to amps, thereby drowning the THD of the amp.
I think this is a misconception, sorry. While it is true that the THD of even the best transducers is typically in the order of 0.25% to 1%, the distortions of commonly used transducers mostly simply add harmonics, otherwise known as overtones, or multiples of main frequency, that do change the timbre/coloration of sound, yet are perceived as natural by the human hearing system.
The solid state amps introduce unnatural intermodulation distortions resulting in appearance of phantom sounds at frequencies (f1-f2) and (f1+f2), and these formulas are only that simple when only two major frequencies are present. With several major frequencies present at any given time in a typical recording (usually 3 to 12), the picture gets a lot more complicated.
Numerous phantom signals at effectively random and constantly shifting frequencies, appearing due to the intermodulation distortion, greatly confuse the human hearing system and make any music sound, well, unmusical. The sound processing neural circuitry tries to filter those frequencies out, thus consuming more neurotransmitters, depletion of which manifests itself as listening fatigue.
The solid state amp's THD is typically correlated with its level of intermodulation distortion. There are good theoretical reasons for that, but it can be also understood on a general level - the more linear and uniform over the frequency range the transmission function of the amp is, the less THD and less intermodulation distortions it will have.
Basically, the most significant source for both harmonic and intermodulation distortions is the same - amp's non-linearity - and it has to be reduced to "unheard of" levels (pun intended
for a solid state amp to sound right. This is an oversimplification, only correct as a first approximation, yet it does indeed depict accurately the dominant factor.
The high linearity is typically achieved with the negative feedback loop in combination with highly symmetrical amp design and very high open-loop amplification coefficient. Achieving this is expensive, as more stages of amplification are required and those stages have to consist of precisely matched components.
This task is much simpler if the amp power requirements are low, as designers can use operational amplifier chips, which achieve high symmetry and ridiculously high open-loop amplification coefficients by the virtue of their on-one-chip design.
That's one of the fundamental reasons why modern professional monitor speakers sound so good - they use multiple transducers, each powered either by a separate high-quality op-amp or a simple circuit incorporating one.
The splitting of original signal onto frequency bands happens before that professional monitor op-amp, using active crossover, so that the op-amp doesn't have to work as hard as the amplification cascades of a typical discrete amp that has to drive multiple transducers through highly lossy passive crossover.
Corollary - a couple of professional monitors that cost slightly over $1,000 may sound as good and as loud as a combination of traditional audiophile-grade discrete amp and elite loudspeakers costing upward of $5,000.
What I wrote is about the unscientific approach of data gathering, your data should be considered feelings and opinions, not proofs.
Yes, of course, those were my opinions. Yet by now you may have noticed that I have some background in electrical engineering and neuro-acoustics as well, so my opinions may have some merit after all
In general, applying scientific approach to data gathering and especially data analysis is sometimes much trickier than it may seem at first. Blind testing using speakers that are behind a heavy cloth is a good example of an experiment where the systematic error is not well-controlled. Thus I don't have much trust in some pieces of what passes as "research" these days.