I have a core 2 duo and have been using 4 threads as the manual suggests.  I will try two threads.  It's all so confusing.  Rendering in a separate process definitely greatly enhances performance.  What about bucket size?  Rendering in a separate process doesn't use adaptive buckets I think I've read.  I had a problem with a IBL/ambience occulsion scene when I cranked the settings up to 100%.  There were terrible artifacts on the image and after about 15 image the picture was only half rendered.