Will we continue to keep the source tarballs in the gate, or will they be external? I know that mercurial has had performance problems with large binary blobs in the past; I don't know if they've been solved. Also note that as of now, there's no way to eradicate history, so all the old versions of things will stick around forever, even long past the time we've stopped caring, and people building current workspaces don't want to download stuff.
Danek
