A few further comments:
- I think when XHTML is pushed as "semantic", the term's used in a relative sense. That is, well-formed, properly structured XHTML is more semantically useful than HTML 3.2 tag soup cluttered with spacer GIFs, table layouts, etc. I don't think anyone ever claimed that it was as semantically rich as, say, MathML. It's a general-purpose document markup language that encourages separation of presentation from structure. Full XML support in major browsers will widen our options, I guess.
- To be fair, even HTML 3.2 can be "properly structured" in the sense used above - though the technology itself doesn't encourage it, as XHTML/CSS does.
- I'm not sure that browsers having built-in presentation defaults (paragraph spacing, heading sizes, etc.) implies that XHTML "centers around visual presentation". XHTML very generally describes hypertext documents. The same one can be read out, or rendered visually on radically different devices (presumably each with their own defaults). The document doesn't change, as, apart from rogue bits like the I and BR elements, the markup has little to do with presentation of any sort.