Yes, these are a nice enhancements, but they are not to the OP's specification which detailed that the statistical frequencies (ie. counts) be stored in the lowest hashes.

I cannot comment as to which is better since I don't have the bigger picture of the problem.