Reading the report [PDF] of the RLG Programs metadata practice survey, this quote from a respondent jumped out at me:
We use a variety of tools to produce a variety of records. Mature and established systems (such as our ILS) are generally effective. Tools for creation of XML are not as efficient - particularly EAD creation. Creation of EAD and ingest into our XML database is still a very manual process. Our tools are also generally not well integrated. Even when describing the same resource we use the ILS for creating MARC, home grown tools for creating EAD, and perhaps a third tool for creating item level descriptive metadata. [RLG Programs Descriptive Metadata Practices Survey Results - PDF]
It is pretty indicative of general issues to emerge. Metadata creation practices are fragmented across different materials workflows with variable systems support.
... RLG Programs surveyed 18 Partner institutions1 in July and August 2007 to obtain a baseline understanding of their current descriptive metadata practices. Although we saw some expected variations in practice across libraries, archives and museums, we were struck by the high levels of customization and local tool development, the limited extent to which tools and practices are, or can be, shared (both within and across institutions), the lack of confidence institutions have in the effectiveness of their tools, and the disconnect between their interest in creating metadata to serve their primary audiences and the inability to serve that audience within the most commonly used discovery systems (such as Google, Yahoo, etc.). PDF]
I was also interested to note that over half the institutions surveyed build and maintain one or more local thesauri.
For more detail see Karen Smith-Yoshimura.