Maybe, though what I was trying to get at with my comment still isn't really addressed. It seems like if you're making everything a filesystem and making it so that the OS doesn't care about where the filesystem is, it can be very easy for latency costs to pile up.

I really should properly play with it, but it always seemed to me that it has the potential to add milliseconds of cost to each operation and that could be very slow.

Just because you can run it over the network doesn't mean you have to. Like, yeah, you can run Linux with root on NFS and yes it can make you vulnerable to latency problems, but you can also run Plan 9 completely on a single machine with all the myriad filesystems coming from from the local system (mostly virtual, but some actually hitting disk).

If you mean that microkernels ping-ponging between kernel and user space can impact perf: Maybe? I'd really want to see benchmarks.