It worked great for me. Peacock, paper, scissors, top hat... it got them all. Top hat did take some 30 questions though. I did my dissertation on expert systems so I am naturally interested in these kind of things.
One day, when this kind of database is large enough (and it will be since all what we put there will be recorded. And hence the nationality and sex demographics) a good performer could put this in a good use.
Think about it. You, as a performer, already know the demographics and store them in. You now have an idea what sort of things people usually think (a starting point). I would say that less than 10 questions should do it. If we add showmanship, observation, and vague enought answers (as Pete said, skates would have produced a miracle hit) ir could seem as real mind reading done with just a couple of questions.
In fact, add a swami gimmick and you can pretend that the last two or three questions for just to let you show off that you even knew beforehand answers to specific questions.
Naturally one would need an assistant but still, it could create wonders.