At a certain point in some use-cases the Linux problems outweigh the "improvements", and more traditional partially conformant posix systems reduce complexity.

For example, the reduced attack surface area of OpenBSD hardware support is a kick in the pants for average users, but desirable for hardened system design.

Why does none of this really matter practically? (seriously it doesn't)

In general, Linux has so many people looking at its code, that the CVE and driver issues will be addressed with higher frequency. Thus, FreeBSD/OpenBSD lower 0-day incident rates tend to be illusionary, as the security incidents in fringe OS always have lower discovery probability.

I am a fan of most things posix, and acknowledge most problems originate from Application space rather than the OS itself. =3