check out how quickly #git-pages (and #Grebedoc) can check out a giant git repository without any changes!
if supported by the server, it retrieves only a single tree from git (no other branches, no tags, no history, no file contents), backfills it from the existing site contents, and then pulls in any missing files from the git server on-demand
this lets you host very large repositories without straining network and compute (for compression, etc) resources!