Which is basically everywhere.
Well-designed microkernels have been proven to have only a minor impact on performance, afaik. Especially if said kernel is small enough to fit in a cpu's L1 or L2 cache entirely.
With that in mind, I'd say there's few cases left where users would not want to take a minor performance hit, if that gets rid of entire classes of bugs & vulnerabilities.
Reasons this isn't the norm these days are mostly historic. But going forward, a good midpoint would be popular OS kernels like Linux split into smaller components, while providing same user-space APIs. Along the lines of how XFree86 was modularized into a set of Xorg libraries. Maybe L4Linux could be an example? (dunno how modular that is though).
On such a componentized system, components could be shared among multiple 'OS personalities'. Kinda like a multi-VM setup but finer grained with much more shared code, much reduced attack surface for individual components, while maintaining very strong isolation guarantees between components.