I'm not very techie, so I could be wrong, but this is info that I've picked up in case it helps.

I have an i7-3930K processor which is very similar to your Xeon; the main advantage I have for rendering is 6 cores/12 threads rather than 4/8. One thread = one bucket, so with CPU rendering the more cores = the more buckets can be rendered simultaneously. The optimum bucket size will vary according to the scene. With CPU's you want to have as many buckets rendering as possible while GPU's render one bucket at a time. Therefore it is best to use the largest bucket possible without running out of VRAM on the GPU, but a smaller bucket on CPU. Complex materials and transparencies could cause a bucket to render far slower than the others so it's possible to be sitting waiting for the render to complete as one bucket is grinding away at one part of the scene (like hair).

The renders I'm doing at present tend to be 800 pixels square and only have simple materials and lights, no transparencies or SSS. I also turn off BPT, use pretty much default settings, bucket size of 40 and 40 samples and I get a final render out in around 15 minutes.

----------------------------------------

Not approved by Scarfolk Council. For more information please reread. Or visit my local shop.