Xeon? Cause you don't need internal graphics when you have discrete. (Unless it is crystalwell, then you definitely do want it)
Dual chip setups are actually actively bad for gaming, when one needs to access the other's cache/ram it isn't good for latency, and you are relying on the os' scheduler software to do the juggle.
My ultimate gaming machine would be quad sli etc, but have a single chip with as many of the fastest cores I could find. Speed of the cores being more important than number(above 4), it would also have the fastest memory I could find.
Dual chips do have a penalty and here is the condensed version:
if a CPU needs to access its own RAM there is no penalty.
If a CPU needs to access RAM on a remote CPU expect waste and added latency, about 1-4% slower
HOWEVER *IF* you have tapped out the QPI (or hyper transport) you basically get the same affect as disk thrashing, expect 20% or more latency (over local speeds.)
Their are ways to combat that, starting with loading one program at a time, the OS will typically assign to the available CPU (and it's ram)
While one program might not use the same quantity of ram modern OS's are pretty good about it.
For the few clients who have maxed out their systems We use AutoNuma for *nix or HyperV for windows (this is great for guys who "bot 10/20 games at the same time)
Even right now I have a couple VM's running on my desktop, and I know that by (my)default all HyperV VMs are using CPU1(and the ram attached to it.)