Huge time variance in rendering

I'm curious if anyone can explain what causes the following kind of time variance, and what possible approaches can be used to help reduce rendering time.

I recently created a small batch of presets on the "Realistic Wood Workshop" filter. I created a batch process to render Diffuse Maps fr om that filter, at 4096px. I then executed the batch, and after 7 days it was only at 86% of the first Diffuse Map. ?!?!

I then did a find/replace on the batch file and changed the 4096s to 2048s, and executed the batch again. Each diffuse map took only 26 minutes to render.

Now, I know that such a process doesn't scale exactly proportionally, but shouldn't a 2048 to 4096 be somewh ere in the neighborhood of 4x the rendering time? 26 minutes to 10080+ minutes is ridiculous.

I should mention that the machine I did this test on is a Windows XP box with 3GB memory, so is this a problem of memory being "filled" and slowing down the process? What approaches would help reduce rendering time? If I were to do this test on a Windows 7 x64 machine with 16GB RAM would the time likely drop? What about the same processor but with 32GB RAM? Would that also help?

Any insights are welcome. I'd also like to hear from anyone who is willing to repeat my test with a 4096px Diffuse Map batch on that specific filter. (I could even provide the exact filter preset to make the test more precise). It would be interesting to see what times we get. I'm going to try the same test on a Windows 7 (x64 4-core 16GB) machine this weekend and see how much difference there is.

Swapping plays a lesser role (though it does contribute to rendering times); what is more important is that Filter Forge writes much temporary data on your disk.

primitivedesign, please check the relevant forum thread, particularly my reply about HDD health.