Multi-threaded processes are used because that's the most efficient model, hardware wise. Tasks can maximally share resources. It's not to do with library design or startup time.