The only use I can think of is if you want to give something a logical id (that is, you don't want to use a database row id for whatever reason) but that is is never accessed as a thing in its own right.

For instance, you might want to identify taxon relationship objects in such a way that other inhabitants of the semantic web can annotate them, but you might not wish to serve up those objects as isolated atoms. Additionally, it may be the case that these relationships in your database do not have stable ids. Or, they might be indexed with a compound key that you do not want o issue some clunky compound URI or LSID for.

To accomplish this, you could give each relationship record a guid which would "follow it around". The relationship records are served up as part of a taxon concept - never on their own, perhaps they might have an rdf:isDefinedBy property pointing to the taxon concept that they are primariliy a part of.

In this case, naming these records with a urn:uuid: guid might make sense. A third party could keep the guid urn, the fact that it isDefinedBy the taxon id, and whatever information that that third party might want to attach to it, and expose that information to the web.

.