There is no magic "one size fits all" setting. Results depend too much on things like footage I/O, the actual processing requirements (heavy effects vs. no effects vs. number of layers vs. blending operations etc.), hardware acceleration getting involved or not down to seemingly trivial things like what CoDecs are used in the source footag. Your results only get "consistent" with MP because pretty much all hardware acceleration stuff reverts to software-only modes and eventualyl you'll hit your machinbe's performance ceiling no matter what goes on in the project - it simply can't get any faster than a certain point. So in short: Stop wasting time trying to come up with magic formulas, tackle this on a per-project basis when needed and accumulate experience.
The biggest problem with the multiprocessor settings and memory assignment is that there are only a few plug-ins that are truly multiprocessor aware. There are also codecs that are not multi processor friendly. If your source footage has an unfriendly codec then turning multi processor on many actually slow the project down.
So I guess it's just as Mylenium said, "There is no one size fits all solution." I never have render multiple frames simultaneously turned on for layout, animation, and set up. Almost all of my motion work is done at half or quarter resolution. If I think a project will benefit by enabling the multiprocessor feature I will run a few tests before I render. Most of the time I just load a comp into the AME and let that pound away while I'm working on the next project. Overall I think that's the most efficient workflow. I don't get paid much for rendering, but I do get paid when I produce.