I like this idea but, as much as this sounds demeaning, I really don't think it will reach enough people that the sample size is going to be significantly large enough to be very accurate.
You should also consider that people on TSR tend to have higher grades than most people, meaning this could inflate these 'predicted' grade boundaries.
Time and time again it has also been proven that we human beings, on the whole, are terrible at accurately guessing how well we did, so these 'predicted grade boundaries' may vary wildly from the actual grade boundaries.
This is adding to the fact that there's nothing stopping people from answering dishonestly, and though this is not likely to be the case, it may also affect the end result.