Just some general thoughts, too many known unknowns and unknown unknowns.

Depending on the render output size you're using, increasing the bucket size could be reducing the number of cores being used - might be an idea to check that all cores are being used by Poser at both smaller and larger bucket sizes.

Another thought is there a performance penalty associated with building a bucket. The CPU doesn't have the CUDA overhead of GPU but it does use the display driver which might make use of CUDA accelerated OpenGL. Try rendering via the queue manager to see if that improves render speed as that doesn't use the display driver.