I would like to second what sandeepss said. 
Where I work we have 80Gb in each full branch on the central server and
developers create local sub-copies usually about 4Gb or so on their machines.
Usually there are about half-dozen of those cub-copies some in different branches.

There is no way people will go for 20 times increase in network delay or disk size.

Most of the data in the repository is tests and test data. We keep those
transactionally consistent with the source code across all branches (i.e. tests
and code always change together) because it makes sense. As a result everything
ends up in one repository.

I am really surprised that neither mercurial nor git support this. It seems like
 a very fundamental feature for me coming from perforce.