The other reply mentioned that threads and processes aren’t much different on Linux. But on some other POSIX systems, privilege separation and resource limits tend to operate on a per-process basis - both of which are important features for a mature RDBMS. If you want to limit the CPU/memory/IO usage of a connection or ensure that a connection never accesses a privileged resource, and you need it done in a portable manner, then each connection needs a dedicated process.