[sandGorgon]

Partial checkouts. I used to work in the semiconductor industry, where HDL Netlists can go upto several GB. If you are in the EDA industry and you need to test your tool on a Netlist or two, you dont want to be checking out the entire test-tree.

I suppose partial checkouts are critical for several other fields - for e.g. take a look at a similar request for Mercurial (http://mercurial.selenic.com/bts/issue105) Message 8102:

Two of the features that made it difficult to 'sell' Hg for the main SCM of FreeBSD were the lack of partial checkouts and partial history

And because of the way legacy code is structured, this gives rise to issues like this one - http://jira.codehaus.org/browse/MRELEASE-457

For several use cases, this is a highly critical feature and the reason that SVN may win over finally - perhaps as SVK (svn's distributed revision control fork)

<empirical-observation> The transition from cvs to svn happened very quickly, to the extent that it is now the predominant SCM. The migration tools are robust today, but they were not so a couple of years back - which is why it is even more suprising that people were willing to undergo the cvs2svn pain but are not willing to do the relatively painless 'git svn clone'. Maybe git truly doesnt suffice in a LOT of situations. </empirical-observation>