Node contents are static. What you type gets stored, thus if you type a cpan:// link, then that gets stored. The parsing of that link into an actual URL happens when you *view* the node.
Oh I didn't really think about that. The thing that had occurred to me was if anyone out there has a script running that checks for updates to a node, that it would see the updated location of the link. My brain was thinking about extraction from the page, but of course anyone doing this would do it via the XML generator, which would simply provide the [cpan://xxxx] text, and not expand the link.