The changes in OOM-722 work fine - we have the same behaviour as before (aaf, vfc issues, periodic sdnc issues requiring a delete/create).
This is running the RI (rancher 1.6.14, docker 17.03, helm 2.6.1 (client/server), Kubernetes 1.8.6 client/server.

The issue is non-standard deployments - the pods up to 19 don't start even after multiple create/delete cycles
env is rancher 1.6.14, docker 17.03, helm 2.8.0 and kubernetes client 1.9.2 (not recommended)
and a kubeadm environment - see comments below