Nathaniel,
the "bible" on interpreting relative locators is the I-D
Relative Uniform Resource Locators
<draft-ietf-uri-relative-url-01.txt>
by R.T. Fielding.
You can find a less rigorous definition in RFC 1630, around page 10.
My thought was to change this mechanism as little as possible
for use within a message. One could imagine the content-ID reference
being done as Content:idvalue (absolute reference), but to my mind,
this introduces too many silly states.
A simple part numbering scheme might also be good, for people who
don't love content-IDs. (I don't generate them, myself)
Harald A