On a Pentium 4 (or Xeon if you are really fancy) with Hyper-Threading it is fine to run with two copies of THINK. With an AMD or Intel Celeron processor, they should only be running one instance of THINK.
Remember that THINK uses only the spare cycles. Running extra instances of THINK doesn't give you more spare cycles. In fact, it will introduce overheads and actually slow things down. THINK depends upon the floating point unit of the processor, more than one instance at any one time will involve them fighting for use of it. Hyper-Threaded processors however, can run two instances due to having an extra virtual core in the processor. Whilst the virtual core doesn't allow two floating point unit instructions at any one time, it will eliminate the overheads that are suffered on the AMD or Celeron processors.
This means that you get maximum performance for THINK using just one instance on AMD or Celeron's and best performance with two instances on Pentium 4s with Hyper-Threading. Remember, two instances will have a lower CPU rating each than running just one instance, but the two combined ought to be higher than the one single one.
What you may have found, is that if you run several instances on an AMD processor, you may have several instances each with the full CPU rating of one instance. However, if you watch the time on the jobs, it will go slower than real time. Overall, this loses you possible points.
As for RAM, more instances will use more of it, but it is the processor that is the most important part here. More RAM is always a good thing for running other programs at the same time. The more the merrier.
I know this post may be a bit technical, for that I'm sorry if it is, but bottom line is - one instance for everything other than Pentium 4's and Xeons with Hyper-Threading. That will give maximum performance.
