I would prefer a site that is aimed at experts, probably at least graduate student or advanced undergraduate, and assuming a reasonable background in theoretical computer science, or a bit of care in phrasing by someone who is not an expert.
Anything below this level seems neither attractive for domain experts to contribute to, nor sufficiently different from Wikipedia or Google Scholar to be valuable.
The TeX site is great because it focuses on information that is not easy to find in an FAQ. It is full of the kind of questions one would ask of the local TeX guru, and the answers are typically those that the most knowledgeable gurus would provide. It is becoming the sort of place where the discussion about a particular question will be the first hit when one asks a related question of a search engine.
It would be good if this site could be the place where we get to ask our theory neighbours (whether across the hall, or on the other side of the world), about niggling folklore questions, tangents that came up while reading a paper, clarifications about new results, or cool new insights that maybe don't quite work. I would also like to see a place where such questions are answered with pointers to non-obvious papers, snippets told by colleagues, historical background that isn't published, and insights that may be common in some parts of the community but which are not universally known.
In other words: I am adept at using a search engine. But I have no access to the corridor gossip at another institution, and almost certainly haven't noticed how the JACM paper you just read is directly relevant to an arXiv paper that appeared yesterday.