Straight up BSD from the CSRG at Berkeley had an absolutely abysmal virtual memory subsystem. Processes had very small and hardcoded VM size caps in the kernel,

Straight up BSD from the CSRG at Berkeley had an absolutely abysmal virtual memory subsystem. Processes had very small and hardcoded VM size caps in the kernel, and some algorithms that worked okay with 1 or 2MB of RAM were scaling poorly with the massive 4 or 8MB systems showing up. Despite being meant as a research system, Mach2 was effectively 4.3BSD with a good, fast, and scalable VM system. I imagine this had to play a role. (The Mach VM would be reabsorbed into 4.4BSD and inherited by all the *BSDs -- it's since been replaced in all AFAIK, but still interesting)