Come to think of it, isn't it supposed to be the purpose of an operating system: letting programs think they own the hardware? Now they can pretend to own the OS too :)
I think we'll get to microkernels, but through evolutionary steps like this rather than ground-up redesign.
Site needs some editing.
Virtualization will be huge in the future, probably we will have common desktop hardware with bare metals hypervisors implemented directly on the hardware in the next 15 years or so, virtualization is already used in datacenters, network appliances and used extensively in the "cloud", how will we use it in the desktop is still open, but I do believe the first application will be running legacy application on modern operating systems and security.
[1]: http://communities.intel.com/community/openportit/vproexpert...
Maybe I need to see something like the Xen presentation at Fosdem again.
Edit: http://www.ok-labs.com/blog/entry/microkernels-vs-hypervisor... is quite ok.
A hypervisor runs multiple Operating Systems, having each one think that it has access to the whole of the hardware.
A microkernel is one way of writing the kernel of the Operating System, so that each part of it is a separate process, routing messages to each other in a safe manner to get things done, rather than doing direct calls to each others code.
With a hypervisor you wouldn't expect each of the OSes to have any communication with each other at all, whereas with a microkernel you'd expect the different processes to talk to each other a lot.
You can, apparently, repurpose a microkernel as a hypervisor, but I don't know anything about that at all. Presumably the infrastructure is quite similar.
http://x86vmm.blogspot.com/2005/11/xen-and-redhat.html
Both present "virtual machines" to client code. For that matter, so does UNIX, although the UNIX virtual machine has some very complex virtual instructions, like fork(2), exec(2), dup(2), etc. You can crudely map most software platforms on a continuum of abstraction, with something like Python's implicit virtual machine (which is dynamically typed and bound late) at one extreme, and a bare-metal VMM whose interface is identical to that of the underlying hardware at the other.
Both paravirtual hypervisors and microkernels extend the underlying hardware, and they do so at a lower level of abstraction than what we call an OS. In practice, the hypervisors extensions feel more like hardware (they might include device models, virtual memory translation, and interrupt models), while the microkernel's would feel more like software (providing RPC mechanisms, security models, abstractions like "thread" and "process", etc.).
Even in practice the line is grey sometimes. L4 used to call itself a microkernel; now it calls itself a hypervisor.
There is a lot of controversy about the distinction between VMMs and microkernels. A view from the microkernel side is given in http://www.ertos.nicta.com.au/publications/papers/Heiser_UL_...
The isolation of drivers in separate VMs and enforced isolation using VT-d is definitely interesting. A paper on the disaggregation of dom0 for improved security is http://www.xen.org/files/xensummit_fall07/22_DerekMurray.pdf