I can't see why you'd want to work at a sampling rate higher than one you're giong to be using for the final product (usually 44100 for CDs).

There is the theoretical aspect of antialiasing, but the differences would be so subtle that I can hardly justify using up twice or four times my current CPU usage on it, and the negligable differnces only become measurable at exceptionally high frequency ranges (I'm talking around 10000htz), let alone audible.

Bit depth on the other hand can certainly be an issue, especially for mixing down bounced tracks. When you compress stuff heavily the artifacting and crossover distortion from poorer quality A/D converters can become very audible on bass frequencies when there's nothing else masking it.