> The idea that microkernels are slow came from analyzing a popular microkernel at the time - mach. It in no way is a true blanket statement for all microkernels.

Don't microkernels inherently require lots of context switches between kernel-space and user-space, which are especially slow in a post-Meltdown/Spectre world? I know that Linux has semi-recently added kTLS and KSMBD to speed up TLS/SMB, and Windows used to implement parts of font rendering and its HTTP server in kernel mode to speed things up too, so this gave me the impression that having more things inside the kernel (== more monolithic) is better for speed. Or is this only the case because of how the Linux/NT kernels are implemented, and doesn't apply to microkernels?