There's one thing that current DVCS really suck at: working with large repositories which are frequent in gamedev and similar areas.
E.g. recently I tried creating a git/bzr/hg repo of 22GB of files (it's not that huge, real repositories that hold all the resources needed to build a game, including branches etc are order of magnitude larger... games themselves are several GBs of *compressed* data, don't forget...).
All three failed.
It seems like all git/bzr/hg assume that every file they're dealing with can be directly loaded to memory. That's probably why git spits "malloc returned NULL" errors when asked to add a file which is larger than amount of memory available.
SHA calculation doesn't help them as well... Running "git status" on even relatively small (~1GB) repository becomes uncomfortably slow.
Yet Perforce handles all this easily.
Subversion guys, do something with it (svn can handle several GB-repos, but it is much slower than Perforce) and you'll have an edge over DVCS!