His opinions as to what bits are significant are as subjective as anyone's, his calculations are not incorrect, and I wish he could have found a more authoritative reference for information theory than Wikipedia. (Or is that an open source community thing?)
I like Bob's conclusion better that we need just a bit more than 16 bits/44.1 Khz. Minimum sampling rate he picks is 56 KHz from what I recall. Bit depth is fine at 16 if proper dither is used. So in reality we want more than that since we don't know if people do the right thing.
I think the best argument for getting high-res music is because that is what is produced. If we can accommodate the bandwidth and storage, we might as well get what was in the studio that the talent approved. Any resampling and bit depth reduction can potentially change the sound some and if done wrong, change it fair bit. For that reason, it is better to not mess with the bits and send them along as is.
