The wikipedia has links to some excellent references and historical perspectives on MP3 and compressed audio in general, as well as some references to results of listening tests. It seems that as you approach 320K, most people cannot discern between the compressed file and a lossless format, but most listeners could tell a difference at lower bit rates. There is also some mention of the possibility that music with a lot of sudden dynamic changes, like well recorded orchestral music, may be more difficult to compress without artifacts, at least in comparison to more dynamically squashed music, like most modern rock and pop. So perhaps the type of music has something to do with perceived differences as well.