Load and availability/security are also factors to consider. Take View or vCloud for example, all the VM operations are putting a HUGE load on vCenter which can cause the service to collapse way before reaching the theoretical maximums.
Also backup software tends to log in for each job and start a propertyCollector iteration which can also put enormous pressure on VC from a performance point of view.
Separation for workload clusters and not having the vCenter managing itself might also be needed in some environments.