A 'system bus' carries data between the processor, I/O, and memory.
The speed of the 'memory bus' CAN BE different than the speed of the 'system bus'. The same is true of I/O bus speeds. It's confusing because you're accustomed to seeing the same speed for both system bus and memory bus.
With a G5 - For tasks not requiring the use of RAM, a fast system bus is, um, faster at moving the data.
PS, don't quote me on this:
I would think that in multiprocessor configurations you would need as much bandwidth as possible in the 'system bus' in order to efficiently link between the CPUs as well as their own memory and I/O busses. Ideally, I suppose, you would want 2X the 'system bus' bandwidth necessary for a single CPU.
Those cheap supercomputer clusters made from multiple PCs are linked together via I/O bus-interfaced expansion cards. The bandwidth available is meager. Real supercomputers have adequate bandwidth between CPUs - and better performance. Look for the bottleneck. It's always there. More is better.