which were caused by issues with Google Cloud Networking, we had been informed by Google:
"A system that is responsible for throttling individual instances in cases where there is networking congestion received data from one of its data sources that were implying that there are multiple choke points on our network. This caused throttling for a few VMs while the network had no real choke points in the path of these instances. There are enough action items on our internal post mortem to detect similar issues faster and not have this exact issue happening again."
We will follow instructions from Google to minimize future recurrence.