I've never had need to get more accurate than chem3D or scifinder can provide. What are you hoping to do?
I think there was a paper a few years back that looked at the various predicting software and compared them to experimentally determined pka's. Might be worth digging up if you are really needing accuracy.