Given the size of the repository, it’s not clear that Git would be significantly better or different.
In all the really big repos I’ve used, a limit gets hit and some wacky customizations are applied. The alternative being that you just have to put up with the sluggishness.
Facebook actually hit git’s limit a while back and contributed patches, etc to Mercurial to work with it. Really interesting stuff. But, stemming from that observation and other experiences, I am a superfan of breaking up repos in DVCS systems. I maintain a mercurial extension to coordinate many repos in a friendlier fashion than hg subrepos (guestrepo!).
I’m kind of persuaded that dvcs is a smell at a stereotypical company though, I think there’s room for an excellent central VCS out there.
I think where we’re heading with Mercurial over the long term is a set of tools that makes doing centralized-model development painless with DVCS tools, while retaining most of the benefits (smaller patches, pushing several in a group, etc) of a DVCS workflow. I don’t think it’s a smell at all.
As for splitting repositories, there are definitely cases where it makes sense, but there’s also a huge benefit to having everything be in one giant repository.
(Disclaimer: I work on source control stuff for a big company, with a focus on Mercurial stuff whenever possible.)
FWIW, I use git with mozilla-central and find it a much more pleasing experience than hg (which I still export to when pushing to shared remote repos). That said, it is also what I am more familiar with, although I did use hg exclusively for a year or so.
I really enjoy having everything in the game repo for many reasons such as the lack of syncing overhead, but it does tend to push performance of version control.
We know, we know… configuring hg is a huge pain. There are plans to fix it.