Okay guys, sofar my testing indicates:
* increasing hash size has no effect on performance
* using a good 64 bit random function and generating the 64 bit hash values all at once is not better than using a bad random function and generating the bits one by one
* using either good or bad random methods to fill the hash values while also increasing the number of bits to 96 did increase performance equally and noticeably.
*going to 96 bits does not increase the storage requirement of the hash very much and barely effects the node rate.
So unless I can actually find a bug in my code somewhere, I am staying with 96 bits for now as it hardly has a cost.