I've run SET@home on my dual processor Xeon with grunty Radeon R9 card, and also my 3770K with another decent GPU.
Yes is chews through the work, but the GPU only shows up as one processor out of the 8 or whatever working on one task. It is faster than the other GPU tasks, but not orders of magnitude.
The work units give to a GPU are packaged differently to those given to a CPU, also each processor is structured differently.
Also the GPU does not reveal each of its internal parallel units like the cores in an individual CPU package.
What you should be finding is that the R9 card (assuming it is something middle of the range or better) will be drawing the same or less power than the pair of Xeons and comfortably crunching the GPU specific.
Please forgive the rushed nature of these posts and lack of links to supporting data as i need to head to the airport.
I love SETI and F@H but my previous research says dont run it on a CPU, too inefficient.
I too love the raspery Pi, and the excellent cousins of it that have popped up, have previously read articls re dist computing on them, can be done, not efficient, pity i don't have a link i can throw.
Checkout the distributed computing forum on overclockers.com.au for more info, or simply post a thread there asking how to get the best bang/$ for SETI
Also on the side, by chance is the Xeon a dual socket 2011 machine?
Facebook recently let go of some server gear and there are Xeon E5-2670's going on ebay for ~ $100 each.
8 physical cores per CPU + hyperthreading and 20MB of cache
Could be a cheap upgrade if your board is compatible?
Gotta run, will read with interest in a day or so.