We compared the numbers we got from our newsletter survey and those we got from Stack Overflow and they were actually much closer than even we expected. Mercurial gets a slight bump in our newsletter responses, but SVN still dominates across the board.
And if you look at questions on Stack Overflow you've got:
532 tagged [Git]
119 tagged [Mercurial]
Just to give you a (unscientific) sense of the bias in the Stack Overflow community.
I think this metric is not really useful: Maybe git is just harder to use (no git/mercurial please, just my first thought on the usage of question tags as a - unscientific - metric).
The problem is the bias inherent in the sample. The Joel on Software development community, Kiln mailing list, and StackOverflow all have something in common - they were started by Joel, and their initial userbase comes largely from followers of his blog.
I wonder what the results would've been had GitHub polled their userbase? Or if you asked a random sample of Googlers (hint: 100% would say "Perforce professionally", and then a good chunk would say "Git personally")? Or if Fortune magazine had had CIOs poll the individual employees of their firms (I bet you'd see a whole lot more Perforce, and a fair bit of "What's version control?")?
Yeah, not really useful. The number of questions needs to be normalized by number and type of people using it, though, for it to be a halfway useful indicator of difficulty. Even then, the amount of information available about each elsewhere on the web will change whether someone needs to ask a question about it.
