The bell-shaped curve gives me more confidence in the result, not less. It shows that Stack Overflow has enough developers that the graph isn't choppy.
The bell curve is a result of the distribution for any population: developers, guitar players, sci-fi readers, etc. In this case, the curve happens to represent SO readers.
You're just looking at the shape of the bell curve. It tells you that it represents a population, but it doesn't tell you what population it actually is. That bell curve could be for the age of people owning cats, for example. There is no evidence that SO users is a good sample of the developers population, in fact it makes sense that it is skewed towards young people that have lots of time to search and answer programming questions on the web.
Stack Overflow is geared towards developers and doesn't have anything I can see to attract a particular kind of developer, except the curious kind. I can't think of a place to get a better sampling of developers right off hand. Certainly, while Hacker News is popular among developers, it would be a less accurate sampling of developers than Stack Overflow, because people here tend to be attracted to startups.
I disagree with what you said in your first post, that it is completely wrong, and that no valid conclusion that can be taken from these numbers. It's not perfect but it's far from being a terrible sample. It's a general-interest developer site rather than a specific-interest one. There are all types of developers. In a reverse-sorted list of popular recent tags, there are c#, javascript, php, java, jquery, .net, and android. Also, developers didn't choose to be a part of this graph; they merely got put into the survey result because they had Stack Overflow accounts. If it had been a survey that was announced on twitter it would be biased towards people who want to take surveys. This would be a worse bias IMO than people who want to ask and answer technical questions.
I wouldn't take issue if you had said it was problematic, but instead you went straight to a one-sided conclusion.
What about developers that don't spend much time online? Or that don't find much personal joy in answering the questions of strangers?
Just because Stack Overflow might be one of the best places to get a sampling online doesn't mean it is actually a correct sampling of the population of software developers.
Those are good points and I agree that there are numerous reasons why it isn't a correct sample. What I'm trying to say is that it's hard to get a correct sample for developers and that if someone is curious about the subject matter it's worth keeping this data until some clearly superior data comes along. There are plenty of datasets that are worse than this, either by having fewer data points or by being even more biased.
