I'm using a custom audio loader protocol in my SNES game because you can do a lot better than the default one. One aspect of this faster loader is that the sound side will just assume the main side is always ready.
Today I went and double checked the timing again and yeah, after the sound side asks for a new set of four bytes, it's slow enough that by the time it actually reads the first byte, the main side is actually about to write the last byte. So waiting for it to be ready is unnecessary.
The audio side of the SNES uses a separate timing source that's less accurate (a PZT ceramic resonator, as opposed to the rest of the hardware using a crystal?) and it drifts a bit with temperature, so you can't rely on exact timing and have to have some leeway. But it definitely looks like there's a whole lot of leeway already.
