fairly recently, my interest in how to design a URI scheme for locations sparked a lengthy discussion on the W3C's URI mailing list, eventually morphing into a discussion of the W3C TAG's httpRange-14 issue (which now lives on as HttpRedirections-57). the heart of these discussions is how the strong foundations about resource identity required by the Semantic Web can be established on the web. Semantic Web technologies assume that identity is established and managed and can be determined by URI, but how that should happen outside of a tightly policed environment is left open.
while the TAG is still struggling with the issue (which is a really tough one), one of the authors of HTTP (roy fielding) voices doubts about the viability of the whole endeavor. the basic problem is that the Semantic Web assumes that somehow people will be able to find out that they are referring to the same concept (because it has the same URI), but in reality careful and managed assignment of URIs happens only within closed user groups. how to bridge this gap between the ideal world of
making statements about well-identified concepts, and the reality that people refer to the same thing using a thousand different names and descriptive terms all the time, is one of the central cruxes of the Semantic Web.
my feeling is that any attempt to turn URIs and HTTP into a technology that serves as the computer network foundation for objective and well-defined identity of all relevant concepts is doomed to fail. there are many philosophical arguments about how easy or even how possible it is to make statements about things per se, and not just about one's view of them. but regardless of which philosophical perspective one is taking, it is pretty unlikely that HTTP anytime soon will be able to solve that problem. my take on this is that HTTP now simply becomes the battleground where the Semantic Web's beautiful theory meets messy reality, and personally, my bets are on reality...