I could be wrong, but my take is that the top-level MIME type relates to
the media presentation/handling capabilities of the receiving system
it's actually both this and default handling. we went through a lot
of this when agonizing over whether to accept model/ as a new top-level.
but IMHO media presentation takes precedence over default handling.
so it's more important for an xml-based audio type to be audio/foo
than xml/foo.