It always amazes me how well class AB amps work when you think about what is going on with the output power transistors, drivers, and voltage amplification stage (VAS).
(1) Current from zero to max
(2) voltage from max/2 to zero to max
(3) non-linear gulps of current from the power lines
(4) Very complex load of speakers, Xover and speaker cable- current and volts can even be out of phase.
This is an area with many views, as we have seen already on this thread, but this is my thinking:
(1.1) Provide the transistors with over-the-top heat sinks
(1.2) Provide as much current capability as you can, at least 2X theoretical minimum, for a given power.
(1.3) Provide as much volts as you can
(2.1) The extreme voltage excursions are very demanding and cause all sorts of problems, including the Early effect, which is a major issue with the voltage amplification stage (VAS), where, by definition, you need to provide voltage gain which implies common emitter operation. A cascode configuration gets around the problem but increases the amps overhead (dropout) voltage. Emitter follower configuration for the output trans and drivers are commonly used for this reason.
(3.1) Use big caps connected to the output trans. Put HF capacitor in same position. Use massively thick wires- multistring or single core- not sure. Use audiophile caps with low ESR etc
(3.2) Use correct-order connections especially earthling and use massive star points.
(4.1) This is the killer because, like with lab power supplies, you never know what the load will be. Having a good voltage and current capability is the first start but getting unconditional frequency stability with such a wild and undefined load is difficult/impossible. And electrostatics look like a huge capacitor with all that that implies in terms of absolute current and the phase between voltages and current.
Being an engineer, I rubbished all the talk in the late 1970s about speaker cables, gold contact connectors, speaker stands and the like, but was proved wrong.
At one time we did some experiments with different speaker cables and there definitely was a difference; the ludicrously expensive Van den Huls sounded the best, but some thick speaker cable from the local store wasn't too bad. 30A single copper-core mains cable was even better. Being a skin-flint, that is what I had- nice grey finish.
Best stop now, but I haven't said anything about the speaker load and cable matching.
Strangely, the LM3886 does not have a symmetrical output stage which is surprising. The rest of the chip looks nice though- perhaps National could't fabricate PNP power trans on the chip.