The random variable `X = # of developers who use OS X` is a binomial distribution (a developer either uses OS X or doesn't, each developer has an equal probability of using OS X [1], and the experiment is the survey). To achieve an error bound of B for a binomial distribution, then we need n = 1/B^2 experiments. Solving for B, we get B = sqrt(1/n). Thus, the error bound of this experiment is theoretically B = sqrt(1/56,000) ~ 0.004.

I'm by no means a stats guru, so correct me if I'm wrong. But that seems like a pretty good error rate to me.

The criticism that StackOverflow isn't a good representation of the general developer community still might be valid. I honestly don't know enough about its demographics to make any educated opinion about that statement.

[1] I make this assumption because there a finite set of jobs. Even if a developer's job requires them to use Windows, they had an equal probability of another developer of having a job that allowed them the freedom to choose their platform/required them to use OS X.