Number of threads and bucketsize influence the speed of the render. Unfortunately this is very much dependent on the scene you are rendering. If there are large "difficult" areas (those where reflection/refraction and/or transpareny are involved) you are better off with smaller bucketsizes and higher number of threads to spread the load evenly on the processor cores. But this goes at an expense - the "not so difficult" portions will go slower overall because it is less efficient.

In my tests I found that a bucketsize of 32 and the number of threads equal to the number of cores is on average the fastest. In practice I set the number of threads one lower as the number of cores to keep the pc responsice to other  tasks. Again - in my tests - on average the speed of P8 and PP2010 is about 10-20 percent faster as PoserPro due to the better bucketmanagement.

Regarding the missing pixels - I have seen them in the early PP2010 beta and lowering the number of threads seem to resolve these.; I have not seen them in the released version. Better open a ticket to Smith Micro with the particulars to get that resolved.