Hey Kelly, long winded rant here, the short answer is having the input stage thermally isolated from the rest of the circuit is probably a big factor. It only self modulates in a discrete topology, whereas in/on a monolithic substrate, a whole symphony of isothermal fronts "slosh" around the chip.
I believe a big blind spot in the engineering community, related to the tendency to measure what is convenient to the current (at the time) SOTA in test equipment, is forgetting that music is dynamic/transient in nature, and test signals are primarily static, or steady state in relation to thermal equilibrium.
Active electronic devices are not just sensitive to electrical stimulation/input (signals and noise), but also to thermal stimulation, and to some degree, mechanical/acoustic stimulation.
Active devices also self modulate, particularly under dynamic conditions, both electrically (load lines, etc) and probably more importantly, thermally.
The temperature of an active device, unless it is specifically (or accidentally) in a constant power topology, will sloppily follow the envelope of a music signal passing through it, as it dissipates less and more power in response to the signal. This is the above mentioned self modulation, as the device's parameters are temperature sensitive, and the mere act of passing a dynamic signal, such as music, results in parametric shifts.
For loop feedback to work, as in track and correct forward signal errors, any wobble or uncertainty in the control mechanism is a big no-no...unlike textbook perfect examples, real world differential input stages look more like mosh pits, having the Vbe or Vgs threshold voltage being constantly perturbed by these thermal modulations. The threshold is in series with the "DC" input offset, and is outside the feedback loop. When passing a steady state test signal, all will be in thermal equilibrium, and look squeaky clean, but under dynamic conditions, again as in music, there will be this wobble offset that is not corrected by the feedback loop, and is made worse by the feedback loop. This is not the same as stable/unstable for the feedback loop, it is a bit more subtle.
For seemingly reasonable amounts of loop gain or "feedback", above 6dB, the feedback signal seen by the differential input stage is larger than the original input signal, making the topology proportionally more sensitive to the thermal perturbations. Throwing output and load characteristics into the mix, as loop gain is increased adds more, er, wobble in the mosh pit.
On a monolithic substrate, as in IC chip, there are thermal shock waves or isothermal fronts bouncing around under dynamic conditions. This is a design issue for high bandwidth, ie GHz and/or high resolution, ie ppm, topologies. These have been simulated and measured, and in typical chip dimensions, the time constants of the thermal fronts and subsequent reflections corresponds to 1/f of frequencies in the audible range. So in/on a chip, the input stage is further thermally perturbed by the rest of the circuit. Unfortunately, low static "DC" offsets in most opamps are designed and measured under conditions of thermal equilibrium, and act quite differently when passing dynamic/musical signals.
This likely correlates with the edgy or fuzzy artifacts some of us claim to hear.
Funny, avionics takes into account "play" or wobble, or hysteresis into feedback and error budgets, but audio usually does not. Maybe because planes crash, but music reproduction just sounds fuzzy or edgy.
Thermal tails/perturbations being made worse by loop gain / "feedback" may well also correlate with why non-feedback designs appeal to some (many?) ears.
Speaking of feedback, it's not the feedback that is evil per se, but the implementation of it that causes sonic issues:
If high loop gain circuits, ie those with "feedback" are stable, actually technically unconditionally stable, then the slowness of the open loop is essentially inconsequential. Feedback is a bit tricky to grok, some of it is counterintuitive indeed.
TIM, in which one of the stages inside a loop "speed clips" or is overloaded by an excessive rate of change in response to signals faster than the open loop bandwidth is generally not an issue these days, particularly at audio frequencies.
It is relevant for opamps used in DAC I/V stages, as the steep edges of the stairstep, or high frequency stuff spewed by noiseshaping can and does wreak havoc inside many poor hapless opamps.
There is a device that has orders of magnitude better thermal characteristics under dynamic conditions: the thermonic valve or firebottle. Another thread...
This rant has not been intended to claim the one true solution (tm) is reduction of thermal tails...for every dogma, there are countless contrary examples, as in tube designs that sound mushy, or products using opamps that sound good, etc. I, for instance, actively practice hypocracy in regards to most dogmas as a matter of principle
and enjoy tubes, monolithic, and discrete solid state designs when in synergistic systems.
There is a phrase, something like: in theory, there is no difference between theory and practice; in practice there is a big difference.
I do believe that if more designers got out of the static mindset, and thought dynamic, including, but not limited to thermal tails, there would be more audio products that have that elusive, magical mojo that is unfortunately uncommon.
