On the HTML WG mailing list, Ben Boyle:
An interesting point, but I think this is highly dependent on whether you are publishing information primarily for browsers. Personally I like a combination of atom and html, both generated from a single source of data (for which I prefer XML). HTML is for the browsers, atom for feed readers and other (potential) consumers/aggregators. I use XHTML within Atom (personal preference to avoid CDATA where I can).
The browsers offer one rendered view of information on the Web among many possibilities. JSON, RDF, Atom, plain text, xhtml, html are parts of the choices to represent an information resource.