Here are a couple of points to consider:
1.) Most aspects of LSP are single threaded in nature. On my i7, the MAXIMUM CPU you'll see while building transitions is 13% (1/8 th of the CPU) as the i7 has 4 hyper-threaded cores (each core appears as 2 to the OS, thus 4x2 =
. For this application, you'll be bound by clock speed, not by the number of cores as only 1 will be in use for most processes. Occasionally during optimization, additional cores would fire up, but it's not all that often.
2.) Most out of memory issues are NOT system limited. The application can only provision roughly 1.25 GB of RAM. Once this ceiling is hit, the application will crash or create glitches. I had 8 GB of RAM this year and one instance of the application running did not TOUCH my available memory.
IF you're going to render / build a LOT of transitions, you can run MULTIPLE INSTANCES of LSP simultaneously. At one point, I had 7 or 8 instances of LSP running and that would consume 100% of the CPU and use 70% of my memory. That really only comes into play if you're doing several very long, processor intensive operations. If you're just sequencing (adding ramps, chases, etc.) then you'll not benefit from running multiple instances of LSP. In my case, I was generating effects for 4166 channels and on average, that would take between 5 - 15 minutes per section. I could easily open multiple isntances of LSP and begin the render process.
If I were building a machine for LSP, I'd look for something with a fast processor, fewer cores (the i3 or i5 series are fine for this) and 8 GB of RAM. I don't know that you need any more, and as cheap as RAM is, I don't know that you save very much by going down to 4 GB. I also don't know that you gain much by jumping to 16 GB unless LSP is significantly revamped to take advantage of multiple GB of memory.
Just my thoughts...