URIs are used in RDF to identify things like properties, classes, and potentially physical objects like people and cars. These things do not fit the conventional notion of web pages, but if they have URIs, we have to wonder: what should a browser do with such a URI?
sandro has suggested using "303 See Other" to redirect from the URI for anything that is not a "web page" to a web page about the thing. Redirects to fragment-addresses often work, but may not be perfect. In any case, it does allow a non-fragment URI to identify an arbitrary thing and, when used in a legacy browser display useful, authoritative information. Content negotiation, META linking, or RDF embedding in HTML can be used to convey RDF information in the same authoritative manner without breaking anything. (The RDF information will have to be written in a 3rd-party style, saying about="http:..../foo" instead of ID="foo", because its address is NOT the leading part of the address of the described object. But that allows RDF and HTML to exist at the same address without conflicting fragment semantics, so it might be a good thing.)