Thanks everyone, especially Denis for the interesting information. Not what I had hoped for but kind of what I had expected. Usability with Pandoc→LaTeX is a must for me. I guess I could write my own code to locate references, look up the data, invoke something which produces a bibliography in HTML, then massage that HTML into something Pandoc will convert into LaTeX appropriately. Sounds like a lot of work though — although not as much as implementing a whole citation processor!
Another question is whether pandoc-citeproc will ignore unknown fields or throw an error? If the former I can have custom fields with the separate original and transliterated fields, as I need for my searching purposes and automatically update the combined version in the official fields. If the latter I'll need to produce a special version for Pandoc's consumption with the custom fields removed, which complicates things considerably.