To address #702, we should try this first, as it should allow citeproc to use the same cache. My only concern is if there's any significant performance penalty, but I'm guessing not.
