I'm making a Connect Four game using the typical minimax + alpha-beta pruning algorithms. I just implemented a Transposition Table, but my tests tell me the TT only helps 17% of the time. By this I mean that 17% of the positions my engine comes across in its calculations can be automatically given a value (due to the position being calculated previously via a different move order).
For most games, is this figure expected? To me it seems very low, and I was optimistically hoping for the TT to speed up my engine by around 50%. It should be noted though that on each turn in the game, I reset my TT (since the evaluation previously assigned to each position is inaccurate due to lower depth back then).
I know that the effectiveness of TT's are largely dependent on the game they're being used for, but any ballparks of how much they speed up common games (chess, go, etc) would be helpful.
EDIT - After running some more tests and adjusting my code, I found that the TT sped up my engine to about 133% (so it took 75% as much time to calculate). This means those 17% nodes were probably fairly high up in the tree, since not having to calculate the evaluation of these 17% sped up things by 33%. This is definitely better, but my question still remains on whether this is roughly expected performance of a typical TT.