It's a theoretical question: So OpenShift V3 is using a cluster-environment with Docker and Kubernetes components. One of this components is a Scheduler - controller manager on the master.
When a new pod is created, the scheduler will search the best available place for that pod, on that moment. The controller manager will check if al the pods are running. When a pod is going down, it will be recreated immediatley by kubelet (not sure?) (this can be on another place). But only pods with a replication controller above it will be recreated. So I don't see the real difference between the controller manager and the Replication controller.