(Semantic web my ass! XML is meaningless; just a container format. :P)
As long as you try to derive the XML from the current wiki markup, it is surely meaningless because it is layout driven. Let us use a well defined schema and you will get the semantic web for free.
As people add "categories" we could also add semantic markup step by step.
I don't think we should muck with the wikitext itself; that should simply be a piece of string data, perhaps in a CDATA section, and the semantics inside that is a matter for another standard entirely. I do, however, want to make sure that the format is extensible in being able to add new kinds of metadata without breaking anything.