Sometime ago I wrote an article (when I had time to write stuff...) on
the various different models of handling multiple databases -
http://www.ariadne.ac.uk/issue23/dovey/intro.html. I think some of the
terminology has changed (I don't think OAI was around then) but I don't
think that the concepts haven't changed.

However OAI does not solve the semantic issues if people use different
"cataloguing" rules. My favourite example is still Russian composers and
transliteration. If one database uses modern transliteration rules (e.g.
Chaikovsky and Rakhmaninov) and another uses "traditional"
transliterations (Tchaikovsky and Rachmaninoff) neither Z39.50/SRW nor
OAI-PMH will solve the problem that the user either needs to be aware of
the differing practices of the databases, or you need intelligent
cross-walking/thesauri software (agreed, I'm sure people could come up
with clever ways of handling these particular transliterations but you
have the same problems in subject terms/taxonomies etc.)

There isn't a single semantic view of the world, and there aren't simple
mappings between different semantic views... 

Matthew