I have to admit that I find more value in just using RDF and RDFS, along with simple to install and admin RDF data stores like Sesame, and perhaps Redland (nice, but does not fully support RDFS inferencing).

Something that works for me is always keeping Sesame running (and OpenCyc, but mostly for learning/experimenting - I have never done anything practical with OpenCyc, after years of playing with it) on a server on my LAN at home. If you always have a RDF data store available, then it is easier to use it as an alternative to a relational database, if that is appropriate for a project.

Anyway, I like "semantic web 'light'" :-)