Postgres forks off a new process for each connection. This introduces lot's of overhead across the board. * Cross process communication is a lot more expensiv

Postgres forks off a new process for each connection. This introduces lot's of overhead across the board. * Cross process communication is a lot more expensive (done via shared memory or as here via the file system) * Switching between processes is a lot more expensive because each process has its own memory space, hence a switch flushes the TLB. Also more bookkeeping for the OS. This is especially bad for a DB, which will usually spend most of its time waiting for IO, so can switch execution context all the time. * Each process also has a distinct set of file descriptors, so those need to be cloned as well * A dB needs lots of locks. Cross process locks are more expensive. * ... These things add up.