Using CPU utilisation as a measure like that is pretty pointless anyway.
Without knowing over what period (instantaneous measure across the farm, a ten minute average from a typical node, a ten minute average over the lot, ...) that reading was taken it could mean many things.
If they have pushed a chunk of the rendering duties client-side then the CPU is not going to be their bottle-neck anyway: the back-end data-store and network bandwidth will be.
Without knowing over what period (instantaneous measure across the farm, a ten minute average from a typical node, a ten minute average over the lot, ...) that reading was taken it could mean many things.
If they have pushed a chunk of the rendering duties client-side then the CPU is not going to be their bottle-neck anyway: the back-end data-store and network bandwidth will be.