In terms of raw compute, you're absolutely right. In terms of understanding distributed computing and the failure modes that can occur, the GP's approach is de

In terms of raw compute, you're absolutely right. In terms of understanding distributed computing and the failure modes that can occur, the GP's approach is definitely better. If you build distributed systems only testing them through virtualisation, you'll run into all sorts of issues that come from adding real world constraints that are otherwise masked through the virtualised abstraction.