Good. While it *can* be worked around, it's a PITA and it risks getting overwritten by other things, since the normal docs loads are based off release tarballs. We can make them off a snapshot tarball, but it's a pain :)
Oh, and +1 for stable links like that in general. That would be one good step.
Not having considered what to do with, but it could also be interesting to teach the loader to read the structured data out of the XML file and store it in one of these weird things called a "database". That could be used for things like matching on the contributors list and add a little per-version badge to which versions they are known to contribute etc. That would require us to be quite consistent in the naming of people, and also not to have duplicates though, but maybe it can be valuable?
Actually, do we have a policy for handling duplicates in the docs there? Since we only keep names there and not email or similar?