You might note that for most schemes, URNs are intended to point to
'intellectual content' which might have a number of different
representations. This is actually quite problematic (who gets to
decide whether two different things should be treated as 'the same'?)
but nonetheless, your third point might be handled as a URN with
possibly more-than-usual variability. What you're presuming is true of
URNs in general (object identity) is really true only of the LIFN
proposal.
Alternatively, your 'keyword' model maps most directly to URCs, which
are either attribute-based identifiers or else attribute-based
descriptions (depending on the context.) So, a URC might include
keywords such that you could invoke a pointer to ("a book by someone
named Melville with Moby in the title") and hope that you might
actually refer to the same book.
The archives for the URI working group are best accessed through
<URL:http://www.ics.uci.edu/pub/ietf/uri/>.
I'm discouraging general discussion on '[email protected]' unless it
directly concerns one of the documents before us, as we have much work
to do, and the entire issue of object identification and distribution
is too large to hope to solve; I don't think any of the issues you've
brought up about object naming and access are specific to VRML,
though.
Larry