Only use OpenGL for accelerating previews. There are certain things and effects that it cannot do and what OpenGL can do depends on your graphics card.
You should think in terms of optimizing your projects and your workflow, rather than optimizing your render settings. As Inlife rightly pointed out, the optimal render setting depends on your hardware set up and on the composition that you are rendering. For instance, transcoding an image sequence will render more slowly with MP, because it uses very little processor. Calculation-heavy effects require a lot of processor. Trial and error until you get a feel for it is best. Personally, I use single processor as a basis (many effects are multi-threaded to use multiple processors anyway, without launching multiple instances of AEs render core… and with less rendercore’s running, each processor can grab more memory). I like to look at what the processors are doing on my machine, then if they’re not fully maxed out, see if MP brings any advantages.
More important in many ways is how you structure and interim-render your project. Look into working with image sequences (targa’s read in particularly fast if you don’t mind being limited to 8 bit per channel). Look also at workflows that employ proxies and pre-renders.
Image sequences are the best, because you can re-render parts of sequences. Have a look at Sequence Destroyer and Immigration over at aescripts.com for more image sequence workflow ideas.
I think that one of the bottle necks for AE is possibly RAM-swapping (i.e. system level virtual memory). I notice that my system drive is often working away like mad when I render something - especially at the point where a render really slows down. I’m beginning to think that replacing my system drive with a fast SSD might bring huge benefits, but I haven’t tried it yet. I’ll report back.
By the way, make sure your disk cache is set to a fast disk too, that isn’t your system disk or the disk with your project on it. And allocate lots of space for disk cache.