Piedpiper,
Great question, which I've asked myself many times. Wish I had a good answer. I was hoping someone would come up with one!
Geoff Husband's theory is one possiblility. He believes tiny VTA/SRA changes are audible due to specific resonance frequency points in the arm/cartridge system. As torque changes it's likely those behaviors would change, requiring adjustment to re-minimize certain resonances. This doesn't really explain everything we hear, but neither does any other theory.
Regarding tonal balance, our reference cartridge does not change tonal balance with changes in arm height. Any ZYX UNIverse owner will tell you that its bass/treble balance do not change with VTA/SRA. Some lower resolution cartridges (like my Shelter 901 for example) do that. I could have used arm height as a tone control with that cartridge if I'd wanted to, though I never did. With a UNIverse it's not even possible .
With a good cartridge what changes with VTA/SRA, in Frank Schroeder's words, is the temporal integraton of fundamentals vs. harmonics. Paul calls it "temporal smearing" (or preferably the lack thereof).
When arm height is just right, each tonal component of a complex note occurs at just the right time relative to the other components. Temporal smearing is reduced and the note sounds integrated. If arm height is off, the fundamental occurs too early or too late relative to the harmonics, making the note sound either fuzzy (HF's too early) or dull (too late). In addition, peak amplitudes are reduced and waveforms are unnaturally extended in time, making each note sound slower and more rounded. This effect is most audible with LF notes.
Whatever we're doing with arm height, it has nothing to do with tone controls. Our ideal setting for an LP reveals more of the music, not less.
FWIW, the magnitude of height changes from one belt to another is not large. From the weakest belt we kept records for (1 mil mylar) to the strongest (2.2 mil mylar), the differential is 32/100ths of a turn on the TriPlanar's dial. That's an arm height change of just .0224". What's notable is the absolute repeatability in both direction and distance. I can't explain it, but I can hear it and repeat it.