Sean Mc Grath is wary of the abstract nature of RDF/Semantic Web. He suggests a "semantic shadow" model of hiding RDF, then mapping individual XML vocabularies to it. I'm not sure that would work.
I think what the Semantic Web crowd needs to do is:
pick 2-3 Customer Scenarios (see Crossing The Chasm) that people can get interested in (one scenario should be non-trivial yet easiest to accomplish, one should be really exciting yet not completely unrealistic).
map at least one reasonable path of Boot Strapping milestones leading to that scenario coming to pass (e.g. who has to play along, what's the business model for them to do so, etc.)