I bet that for every case of scientific fraud that is obvious from the published paper (like this one), there are 10 cases of scientific fraud which are never detected.

Think about it - a domain expert will do a far better job of faking data than a random joe, and will be aware of most statistical tests that could find out.