The trouble is that I make this statement from a place of intuition, conscious of my own biases, so I could be wrong.
I don't think it very accurately reflects what's used in real life, either for the innumerable hours of boring everyday programming work that people do or even what kind of code is being cranked out in the major open source projects - the ones that are actually used by other people.
What it does provide is an interesting leading indicator: what's hot right now. Languages like clojure which get a lot of attention for a while but aren't used in a whole lot of real applications tend to peak and then drop relatively quickly.
As you said, this is still interesting, even if not accurate in absolute terms, as an indicator of what technologies are currently emerging among passionate developers.
That's a bit oversimplified IMO. SO doesn't reflect "interest in" as much as "having difficulties with". There must be "WAT"-type effects vs. simple, intuitive languages like Go factored in for it to provide useful information regarding popularity. It's a bit like counting hospitalized patients with sports injuries to rank sports activities by popularity.
As for the relevance of github - people are always claiming their language detection is broken and I'd worry about the (intentionally?) skewing effects of moving the core language/library development to github.com like the Go project did.
They are using it to have difficulties, so I think it is an interest in. Also, perhaps those that are beginners in things might have more questions, and show that also shows interest.
This redmonk has some horrible graphs, and as you said, the method at which they categorize on github is flawed, let alone using stackoverflow or github as indicators being flawed.
Which obviously stands in contrast to the reality that there are many enterprise systems - some of them mission critical - still running on the platform.
But from our perspective, that tells us less about what might be used moving forward than what is actively being discussed and written in by developers on the platforms we survey.
So no, we would not argue that these rankings accurately reflect the current distribution of languages within large enterprises. We do feel that the actions of a large number of developers from two different properties has the opportunity to be more predictive, however.
Anything other than C and Assembly is sacrilege.