Hello fellow blenderers!
I just unwrapped my new and shiny iMac 2,8 extreme and did some rendering tests. It seems that you can render faster if you set the threads to 4.
I repeated one rendering few times and here is my results:
10.27 - 2 threads
9.44 - 4 threads
10.34 - 2 threads
09.75 - 4 threads
those are like 10 seconds and 27 milliseconds or something.
So is there 4 cores in the new iMac extreme 
try 5 or 6 threads…
Often even putting multiple threads on a single core without hyperthreading gets me faster renders. 2 threads IMO should be the default these days in the b.blend file.
I have an old one core cpu and it renders with two threads faster then with one. I would guess it has something to do with the cpu receiving object and material information before it begins to calculate lighting.
Hum, I’d say it’s because the operating system gives more time CPU time to blender since two threads are running for one same app. More total GPU time = more data processed per ordonnancer cycle = faster!
Dani
Dani is right, I would say the same. The background processes get less processor time when there are two or more threads in the blender process. But I think it is better to increase the blender process priority level, than just increase the number of threads it has to generate. But the speed gained by any of these techniques won’t be enormous, I would just end any unnecessary processes.
Yeh, multithreading doesn’t necessarily mean multiprocessors. For example the Pentium 4s with HyperThreading are one CPU core, but support multiple threads.
Your iMac has a 2.8ghz core 2 extreme, which is 2 cores. A powerful processor, if a little expensive for my tastes 
The only iMacs with 4 cores are the mac pros, which also weigh in at a hefty sum.
The CPU will often have to stall out when waiting for the Hard Drive, video access, etc. since the buses to these devices are much slower than the internal clock of the CPU. Thus it does help to have multiple threads per core since while the one core is stalled, the other can keep working. When compiling applications, it is recommended that the number of jobs be the number of cores in the system + 1. That way all the cores can stay busy.
The Sun Niagra chip takes this all a step further by allowing threads that are waiting on the memory to hand over execution to another task. But I don’t think PCs can do that yet.