That's a strawman. You in fact don't need a "universal ontology," you just need people to agree on first principles (e.g. URLs are unique, there are things called triples, etc.)
I think you do need a universal ontology if you want to make the kind of progress the semantic web people talk about. If you just have a bunch of small, separately created ontologies, the situation can indeed seem great until each expands. Then the intersections and ambiguities become huge.
Sure, if you weren't concerned exactness and lack of ambiguity, you could expand the world of triples into a giant, poorly organized collection of information. It would be kind of like the web. The approach "works" but we, uh, already have the web.
Also, the Doctorow document excellent. Anyone expected naive metadata to be extensible should have a reply to it.
If you just have a bunch of small, separately created ontologies, the situation can indeed seem great until each expands. Then the intersections and ambiguities become huge.
