I want to implement custom scheduling semantics for a cluster running on Google Container Engine (specifically, I'd like to make use of MostRequestedPriority that was added circa 1.5)

I understand from here that one cannot configure GKE's kube-schedule, as it is out-of-reach.

My next guess was to try deploying multiple Kubernetes schedulers, but following the guide, pods that request the second scheduler don't get scheduled, and the scheduler pod warns:

2017-02-06T20:59:57.653394031Z W0206 20:59:57.653076 1 client_config.go:481] Neither --kubeconfig nor --master was specified.  Using the inClusterConfig.  This might not work.
2017-02-06T20:59:57.710633908Z I0206 20:59:57.710463 1 leaderelection.go:247] lock is held by gke-658313547898-12d2ae9e593b23f9af1a and has not yet expired
2017-02-06T21:00:01.165367631Z I0206 21:00:01.165169 1 leaderelection.go:247] lock is held by gke-658313547898-12d2ae9e593b23f9af1a and has not yet expired
...

Is this currently possible at all, and if so, how should I approach it?