I am a bit confused ... you could do RDF without giving URIs any
special status, but that would be like doing a logic without symbols
for individuals.

URIs in RDF are very different from urls for the web. On the web,
they specify the address of a document (more or less). In RDF, it
is possible that the URI interpreted as an address may provide some
information about that object, but that is a side bonus. For RDF,
URIs are just (namespace qualified) symbols.

Given a resource, we do have the issue of figuring out who might have
information about that resource. Various schemes, ranging from approaches
that resemble DNS (my preferred) to freenet are possible. However, I
doubt whether dereferencing the uri for that resource should/will play a
big role in this.

 This is not to say that the central role URIs play in the semantic web 
is deserved. On the html web, every url has a unique "owner" --- the
person publishing the document. There is no such thing in the space of
resources. Who  is the "owner" of Gauss or Oxygen or  ... so we tend to
make up uris. This leads to the rather ugly but very real problem of
different people referring to the same object using different names.
This is a fundamental problem, one that is encountered as soon as
we start building realistically large systems using rdf. (And no, being
able to state equivalence between uris is not the solution --- it is simply
a way of writing down the answer once it has been found).