RE: Using fragment identifiers with URNs

>If it were possible to have a URN scheme for
>which retrieval was not meaningful (and I still haven't seen any
>official document that answers this one way or another) ...

I think "retrieval" isn't the right word, since what you do
with "mailto" or "tel" isn't "retrieval" (HTTP GET) but some
other kind of access/connection. 

Once you generalize the mechanism-of-connection and you
allow 'think about' as a connection method, then of course,
there are no URI schemes/URN schemes that don't have a
meaningful connection method.

(Consider 'tdb' in draft-masinter-dated-uri-00.txt).
 

> Other that what is in RFC 2396, I don't think one can
> meaningfully talk about an "IETF notion of a URI".

There are a large number of IETF RFCs that describe
and constrain URIs besides 2396, and they surely belong
in the category of documenting the "IETF notion of a URI".

At a minimum, I'd include 1736 and 1737, 2717 and 2718,
2141 and 2611  as well as the many RFCs that define and
register schemes and namespaces.