Thanks to all who commented to my previous post, it's made me rethink and clarify my position on the problems with scaling Semantic Web technologies. I boiled it down to this:
Semantic Web clients beware: URIs are syntactically universal, not semantically universal.
The rationale for this is that although it is practically impossible for two disconnected parties to 'mint' the same syntactic URI, it is much more likely that two connected parties will use the same URI to refer to two similar but differing concepts.
The conceptual difference may just be missing qualification: e.g. one document refering to somebody in 1995 and another in 2001. Or it could be genuine misunderstanding: For example our unix server team includes the location of a server as part of its identity - when a server is moved to a different datacentre it effectively becomes a completely different server. Other teams assuming they understand the identity scheme get a surprise when the linked unix server data disappears.
Now this isn't a problem with URIs per se: any other universal identifier scheme would exhibit the same trait. It is however a problem for Semantic Web software which commonly treats URIs as semantically universal without qualification. The bottom line is that before merging RDF graphs one must first manually compare the contexts of each source document to confirm that the URIs in them are mutually compatible.