Surely a clean-sheet design started _now_ would be more interesting and perhaps even useful than a 21-year-old microkernel. Think about all the changes in architecture, networks, storage, etc.
We're in a world with consumer processors with 4-8 cores and hardware multithreading, GPUs that rival or exceed (in certain domains) the processing grunt of the CPUs, network cards that can talk to L2 cache, SSDs, etc.
Surely enough has changed that a clean-sheet design could be a lot more exciting - and even more practical - than trying to finish a design that hasn't succeeded for 2 decades. A clean-sheet design would at least be interesting - we already know you can build a workable system on top of a microkernel.
This isn't a claim to know what a clean sheet design would look like; I haven't really looked seriously at OS research in 15 years. I just strongly suspect you might do radically different, interesting designs in 2011 vs 1990.