I believe most of them still require sizable amounts of work. For example, this individual comment might be able to narrow me down to native U.S. English speaker with higher education, but not much beyond that.
I have played with gender-identification software[1], which has such a limit. Frankly, the identification seems to be more topically sensitive than it is gender sensitive. For example, if I put in my writing about robots it will id me as male, but if I put in my writing about cooking it will id me as female.
Combining the 'author identification' software with known identities (such as we have here on HN) will certainly make them more accurate. Hence, my desire for a system that limits such computations.