Some attempts are using decreased number of nodes in tests, e.g.:

• In (green) https://k8s-gubernator.appspot.com/build/kubernetes-jenkins/logs/ci-kubernetes-e2e-gce-scale-performance/192 I see "Scheduling additional 3883 Pods to measure startup latencies"
• In (red) https://k8s-gubernator.appspot.com/build/kubernetes-jenkins/logs/ci-kubernetes-e2e-gce-scale-performance/195 I see "Scheduling additional 2175 Pods to measure startup latencies"

I'm not sure about the impact of this: whether all the test is using lower number of nodes or just latency measurements, but this looks suspicious and requires investigation.

/assign
/cc @shyamjvs
/sig scalability