David Gerard wrote:
"Imagine if Wikipedia could be mirrored locally, run on a local mirror, where content was pushed and pulled, GPG-Digitally-signed; content shared via peer-to-peer instead of overloading the Wikipedia servers."
The idea of P2P distribution is good. The idea of using Git for this is not.
This is exactly what Git is *not* optimized for: lots of individual files with almost no relation to each other. The same reason you are advised to not use Git to version control your home directory.
Git handles trees, not individual files. In a wiki like MediaWiki, each article has its own history and revision control. Merges to an article doesn't mean that the whole tree (many gigabytes big) must be handled in the same tree-wide commit. The user must be able to commit changes to the most recent revision of, for example, [[Los Angeles, California]], and push such commits while still holding an outdated revision of some unrelated article, like [[Comic opera]].
The rate of changes in en.Wikipedia can me measured in edits per second, and these edits are only related to each other (ancestor-descendant relationship) in each own article. To move to a model where the whole Wikipedia is a single repository with single tree-wide revisions would severely disrupt its efficiency.
Ironically, the per-file revision control model employed by now-obsolescent VCSes like CVS and RCS would fit Wikipedia better than Git (emphasis on revision control *model*, not software).