> Today's main problem for toy and research OSes is hardware support, so I see the potential tendency of them running well on hypervisors just like on the IBM 390 decades ago. But once you do that you kind of miss the point of running an alternative OS in the first place,
With IOMMU visualization like Intel's Vt-d one can run a toy OS against a single piece of real hardware like a network card while the rest will be provided by a hypervisor. That can bear very useful results like isolation of increasingly complex network drivers and protocols behind a hardened special-purpose OS. This reduces the attack surface against other software running in the hypervisor.