Tripdad, this
old thread answers your question I think.
Exactly the same sort of question starts off the thread: "Hey, my PT database says 40% of people win, what's up with the 10% estimates I hear about?" About ten posts down I chime in with some statistical and simulation analysis to show exactly why the PT number can be so misleading. Basically it's the dread problem of small sample sizes once again.
Hope you find the older posts useful.