I'm not misunderstanding the information theory, but I may be misunderstanding the electrical/mechanical context. Does audio equipment use off-the-shelf microchips that are 32-bits, or does every extra bit require more circuitry? I was assuming this stuff is based on computer chips and thus 32 bits is natural--I was just arguing that it's harmless.
Because if your composite data is some infinite-precision data (signal) plus some low-magnitude data, clipping the data does not preferentially remove the noise. It simply changes the low bits to 0, which may or may not agree with the signal. Mostly what got me about your post was the idea that these extra bits are like adding a RNG to the signal. Well, that's true, but so is deleting them.
If making a 32-bit system is harder and more expensive than making a lower-bit system, then I'm fully with you. Either way, charging more for such a system is predatory.