More isn't always better.

Every rendering operation has parts that are not parallel (bookeeping operations, building scene database, etc.)....at some point rendering time is dominated by non-parallel tasks so there will be little further benefit from adding more processors.

...Parallel efficiency is defined as the speedup divided by the total number of processors. On an archaic computer with eight processors, typical scenes render about 7.7 times as fast as a single processor, so the efficiency is about 96%.

....As an example, thread parallelism works very well for small numbers of CPUs such as eight, but 32 CPUs only give a speedup factor of 24. Probable causes include data duplication and system transfer overhead, and the point where adding more CPUs reduces performance is quickly reached because too much time is spent on overhead.

...Finding the optimal number of (threads) is a matter of testing. Scenes with large amounts of scene data (polys) but simple and fast shaders will work better with fewer machines, while scenes with simple geometry but complex shaders such as fur and volumetric effects will work well with larger numbers of (threads). As a ballpark measure, begin with three or four (threads) and do not expect to efficiently use more than ten on a typical low speed network.

Paraphrased from the Rendering with mental ray, 3rd edition chapter 18 Thread and Network Parallelism.