I've been looking at a similar situation and have read up on the support forum about this.
Unlike single core CPUs, if you have HT enabled on your P4, you can run 1 CPU client and 1GPU client (with your X1600) without a dramatic loss of performance. The X1600 should do roughtly around 200 PPD (OCing helps a bit). The CPU client will vary around 150 PPD for non-bonus projects and more for bonus projects. Running 2 CPU clients instead of 1 CPU client would only give about another 20PPD (as it is just 2 virtual cores). So the GPU client is a much better choice for the second client.
This is one of those times that HT helps almost as much as a dual core, as the GPU client appears to use up a core, but it doesn't really do much computation so a virtual core works almost as well as a real one.
BTW for the X1600 it doesn't make much difference if it is the Pro or XT or even 1650 Pro version as GPU memory speed doesn't make much difference - GPU speed & # of shaders is what is important.
p.s. cool to see that the DH folding subteam has its own forum area!
edit: yeah I know this thread is a bit over a week old, but just in case you didn't try it, it probably is worth a shot.
As a side note I've read that some people have complained that their system 2D GUI is sluggish when the GPU client runs - the impact really depends on what system you have and what apps are running so alway worth a try. Otherwise let it run only when you are not using your system.