The basic point, as far as I can tell, is to decentralize regular data--that is, instead of having to go to eBay to list an item for sale, you list it on your page, and eBay simply spiders it, picks it up as "for sale", and lists it. Instead of keeping your own private database, you'd basically be using the entire Internet as your dataset.

Any uses I thought of except that one were just as easily accomplished by having a public API to your web application, which nearly serves the same purpose. The only difference is that the semantic markup formats are "true standards" for accessing the data thus contained, rather than ad-hoc specifying one for every application you create.

Running a myspace/facebook "list of facts/preferences" style profile on your own webserver, that is spidered by someone like Google, and becomes queryable in the same manner as it is on those sites. You could even have a cryptographically signed token from some trusted entity that verifies that you legitimately have a particular .edu email address embedded as a semantic element on your page, and could use a spider service from google to collect comments made from your "friends" to put on your "wall", verifying their identity without them having to log on to your site. The one thing you can't handle under this scheme is varying privacy levels.

Example: You're looking for a house.

There may be several databases of houses for sale in the area you're looking. You want one that's within walking distance of a good school; the regional government has a database of schools and data about them.

If those databases are published in RDF (or some machine-readable format that can be transformed into RDF), then it's trivial to smoosh them together and query them as one.