Why is it that linear power supplies are still used in audio amplifiers for the domestic market? Why not forgo heavy, bulky, old-fashioned caps and toroid transformers for a modern switching power supply that fits into half the rack space? Many manufacturers of high-end commercial amps take this approach.

Basically noise. Switching power supplies are very noisy (electrically). Linear supplies are much quieter. In commercial environments the noise is usually not a factor as ambient conditions have a higher noise floor, and the sound source is farther away from the person. In the home the noise floor is usually lower and the speakers closer. So you would hear the noise throught the speakers at home which you would not in a commercial environment.
It also has to do with specmanship. Because they are quieter, use of linear supplies produce better noise specifications for the products they are used in.
This assumes however that both supplies are designed properly. A poorly designed linear will be much worse than a well designed switcher.
There is no free lunch, you get better efficiency with a switcher, at the expense of noise, and in some cases stability.
In the precision analog electronics I design for medical use, linear is still king. Switchers work for digital and places where we can live with the noise.

My reference to commercial amps might be misleading; I'm not talking about a Class D amp, just a switching power supply with a typical AB or whatever class of amp is used in domestic equipment. Although that might be an interesting discussion too...

Wouldn't the noise from a switching power supply, running over 150kHz, be completely inaudible even if it heterodyned with the 96kHz sample clock of some digital sources?

I was also referring only to a switching supply in an AB environment. The problem is that even at 150K, I have found noise components at audible frequencies due to beat frequencies, bad filtering (ringing) etc. It also introduces harmonics on the grounds which permeate the entire system, resulting in a general high energy noisy environment, which can show up at the outputs as white noise.
I'm not saying it can't be done, just it would be very difficult to get the kind of performance a home user would demand. In my opinion I think thats why class G was created, to give some efficiency with a reasonable noise expectation.