Interesting, please correct me if I'm wrong, but I thought that even though this is true, but since the microkernel architecture is more robust, you can put a lot more "stuff" in kernel space, and you have fewer daemons running in userspace. Naturally this makes things that fail and would take down a monolithic kernel, not suitable for kernel space, but since you're micro, it's now okay.

So overall you have less need for context switches for build in kernel services, and maybe it works out performance unless you're running tons of userspace code.

(I hope I'm not spreading myths!)