This is more of a note to self than anything else, so bear with me.
The idea is to write a module that would speak Z39.50 to research library catalogs, and that would export a PatREST interface to the public. In this way rather than having to deal with either meticulous screen scraping or horribly binary encoded formats you could create a simple client (like John Blyberg's award-winning gadget) that would talk to a wide range of very large library catalogs including the Library of Congress.
A variant on this theme would be SRU REST, which would start from SRU/SRW and export PatREST on the user side.
I think you could build proof of concept versions of these using existing toolkits and only a modest amount of work in an evening, though for many of the neat parts of PatREST (like "hot books" and recent additions to the catalog) I don't know if either Z39.50 or SRU/SRW have those concepts in the protocols.
(More later on this with links to everything, but I had to write it down before I forgot it.)
UPDATE 9/22/06: Richard Wallis does an across-the-pond take on this idea with his Zee REST observations, noting that the Yaz Proxy from Index Data takes care of the Z39.50 to SRU step. There is the question of why not use SRU directly, and the answer seems to boil down to client-side complexity.