I talk a bit about this here: https://codesandbox.io/blog/cloning-microvms-using-userfault.... Before VM A updates its data, the data is copied over to VM B if VM B hadn't written/read that data yet.