Let me ask a sort of oblique question: Do you think recording artists use 16-bit, 44100 Hz sampling when recording and mixing a song? No, of course not. It's best to leave the downsampling to the very last step. I think the same idea applies here. We're all trying to make silk purses out of sows' ears, and one of the best ways to ensure you're not losing any of the already-marginal quality available is to exceed the specs of the incoming signal. Oversample, if you will. Processing can take place in a format with more fidelity than what's available from the source or will be present in the output, which means you can fiddle with things without much fear that the output will have gone through yet another downgrade between capture and rendering. I think that's all that we're trying to do here.