(All of this information, and much more is available in Duncan Epping and Frank Denneman's awesome vSphere Clustering Deep Dive book)
Resource pools don't only take effect during contention. If a resource pool or VM is configured with a limit the available resources may be used up before the host or clusters resources are consumed. Using a limit can be a good way to play around with resource contention without overloading a host.
Contention will occur when resource demand exceeds available resources. This would be when CPU active and Memory active exceed available CPU and/or memory on the host (this would be less than 100% as the host requires certain resources to operate). The reservation and share resource allocation settings (on either the VM or the resource pool it is a part of) affect entitlement.
Does this answer your question?