I'm a big fan of Git. I switched DarkRadiant to Git years ago, and have been happily using it and sharing commits between my and Greebo's repositories (on entirely separate hosting platforms) since then. But even as a Git fanboy I am embarrassed by some of the misconceptions about, both in the "for" and "against" camps.
Misconception: "Git doesn't use a central server".
Reality: The vast majority of Git projects, and the workflow that would most likely be adopted by TDM if it were to switch to Git, is to have a central authoritative server that every trusted developer pushes their changes to. The Linux kernel is a notable exception, but is also very atypical in both its size and the distributed nature of its development. Git provides a little more flexibility in that developers can make a whole string of commits locally and then push them in one go, but they are still going to be pushed to that central server before anyone else can use them. Having each individual developer push their changes to their own GitHub repository and submit a pull request would be possible, but probably wouldn't solve any problems for this project, and would increase workload for whoever is appointed to review and accept all of the pull requests.
Since all developers would be pushing to the same repository, code breakage is just as possible with Git also.
Misconception: "Git magically avoids code conflicts".
Reality: Code conflicts are the result of divergent, incompatible text changes in the code itself. If you and I edit the same line of code and then try to push our changes, a merge conflict will happen. The choice of VCS makes absolutely no difference to whether this conflict will occur, although Git does provide slightly better tools for dealing with the conflict when it does arrive.
There are only two repositories that matter for the Dark Mod: the source code repository, and the assets repository (which is never going to be switched to Git because it is a repository of large binary files which would not be handled well by any distributed VCS tool). So needing to clone the whole thing is not an issue.
Those are both workflow or cultural issues which cannot be solved by a VCS. People can just as easily delay commits/pushes with Git, and they can easily make lots of small incremental commits using Subversion. If commit frequency is a problem (and I have no reason to believe it is), it needs a social solution, not a technical one.
On the other side of the fence:
It does have unsafe ways to edit history, but they are not particularly "encouraged". Most Git guides I have seen make clear that rewriting history is very dangerous, completely screws up other people's work, and should be avoided whenever possible. Sites like GitHub even have the ability to administratively disable force-pushes altogether to stop people from doing them accidentally.
Misconception: "Git adds a whole load of distributed functionality making it hard to use".
Reality: For the most common use case (pushing to a single central server), there is absolutely no need for any Git user to deal with "distributed" functionality. You can produce a SVN-to-Git cheat sheet that maps commands almost one-to-one, and stick to those commands entirely. Some of the commands aren't particularly well-named (e.g. using "checkout" to switch to a branch, or "add" to add changes to a file rather than adding a new file), but they are not complex to use, and you can always set up aliases if you find the default names confusing. Conversely, many operations are significantly easier (and faster) in Git than in SVN. Only recently did SVN add the ability to view the diffs of each change in the history, or to show which changes were pulled in by a merge, and these features are still slow and cumbersome, whereas Git (and Hg) did all of this from the beginning.