Hmm.... that's a tricky question. I would certainly stay away from
using the %Y %B %d etc. notation. That would result in a weird
combination of languages. The XPath notation is I think somewhat
easier to read and use than using the XML way. I think I would go
with the XPath way, as that seems to be easiest for the current
citeproc implementation, and any future implementations in Perl etc.
should be able to convert [Y] to %Y etc.