Very few people add RDF (etc.) to their web sites, preferring to spend effort creating "stylish" looking web sites to creating web sites useful for both software agents and human readers. I am a little guilty of this myself: I have some RDF data on my main site, but it does not reflect the real semantics of my site.
The problem is that when adding new information to a web site, ideally there would be some "standard" OWL ontology that perfectly reflected the subject matter for the new material and an authoring tool would both help "fill in the blanks" to create an instance and provide choices for XSLT to render this instance to HTML. Perhaps an extension to the Protégé system could do this. The problem is still that few people would want to think in terms of what is really knowledge engineering just to add material to web sites.
What is needed is a set of tools that cover all of the bases here: authoring tools that allow fast selection of an appropriate ontology and then instance creation. Then, there would have to be a light weight framework (perhaps server side PHP) to render HTML from an OWL instance and the author's preference for XSLT (assuming that each OWL class also had several available XSLT styles available in the authoring tool -- I am really wishing here!).
Anyway, none of this will happen soon, but I wish that it would. Authors will only want to enter information one time so I am skeptical of systems that have a content creator write content for human readers, then separately try to play nice by supplying RDF, RDFS, OWL, etc.
A "holy grail" of AI research would be to solve the hugely difficult problem of knowledge acquisition directly from text - let's not hold our collective breaths waiting for that to happen: I have been doing some personal research trying to write highly tailored semantic extraction code for a very simple OWL ontology. What a feeling of deja vu: reminds me of my efforts in the 1980s to write conceptual dependency (CD) parsers.