You are using full resources. 100/8 = 12.5 which gets rounded to 12 (you have 8 hyper threads). See NOT ALL tasks are inherently paralellizable. So for some things indeed the speed of one thread is maximum. See the problem is that this kind of simulation is inherently serial in nature (one thing happens after another, so the other processors would need to wait for the one to finish, jsut slowing you donw unless the guess of spreading problem is right). Which means it can not be very efficiently and generally fully paralellized.
Certainly tough rigid body collision calculations detection could be somewhat paralellized but you still not get the full power of your machine, and mayas rigid body engine is certainly NOT engineered to do this.
I'm only using 2.8 GB of RAM out of 8.
Same thing. Problems do not necessarily need more memory to be solved at their peak speed if you intendt to use maya for the simulation. Even so most problems dont get faster with more memory. You see making the computer 10 times faster does not allow you to scale up complexity 10 times for all problems. Indeed we still have groups of problems that can not be solved any batter with today's computers than a generation ago, cryptography builds on this.
As I'm sure you can imagine it's taking me forever to get any results with this type of performance.
And theres nothing you can do about this.
See theres a reason why we get the most number of processors + the fastest individusal core at the same time. For most users out there indeed the multicore systems did nothing for their performance. Most stuff just does not multithread! Rendering out the images just happens to be one of the few well known things that can be easily split into totally independent parts.
PS: you could use nucleus for this its MUCH faster and more modern.
PPS: even if using all 8 cores optimally youd only speed things up by a factor of 4-8 (hyperthreads are not real processors) so say 6 that means a 10 minute job becoes a 2 minute job. Still not stellar performance. So start setting up collision layers etc. And no it doesnt mean you could tackle 6 times bigger problems. more like 1.2 times bigger ones.