The CC-Interop project has been looking at the articulation of physical union catalogs and federated catalog approaches. A conference report and set of conference slides have appeared. From the discussion of Z39.50:
There still exist considerable variations between all of the systems tested in their support for Bib-1 attributes  to the extent that there was difficulty in finding attribute combinations that were shared by all of the tested systems. Exacerbating these difficulties was Z-server default behaviour where query attributes were ignored or replaced due to their being unsupported. The solution to these difficulties is the vendor support of a common suite of Bib-1 attribute combinations such as the Bath Profile [ [‘Hyper Clumps, Mini Clumps and National Catalogues: Resource Discovery for the 21st Century’, Ariadne Issue 42]
The project final report is also available. It reports good technical interoperability, but lower semantic interoperability given variety of practices between target sites. It also notes the user expectation for Google or Amazon style of searching. Although it was beyond the scope of the study, it would have been interesting to know what technical and business impediments there might be to harvesting catalog collections into COPAC, the physical union catalog in the study. Similarly, it would also have been interesting to explore whether linking from Amazon, WorldCat, RedLightGreen, COPAC, or other large bibliographic services into the individual catalogs would meet user expectation for discovery and location better than the configurations being investigated.
There are also some findings about the use of collection level descriptions.