Is the overhead for thread switching in Python greater than for C/C++ threads?

For the most part, threads in the C Python implementation must obtain a lock on the interpreter to start executing bytecode. So I'd expect that Python threads have a strictly greater context switch time than the userspace threads on which they are based.

CPython uses native OS threads when you create new threads using the threading module. Every thread you make maps cleanly to a native thread. However, as the talk in Dmitry's post points out, Python will end up competing with the OS a great deal, and this competition makes thread switching in Python expensive, especially in some pathological cases involving fun things like signal handling (although some recent patches have fixed some of these issues). Essentially what happens is that the OS's scheduler may want to run a thread, but even if the OS schedules a thread, it doesn't mean that CPython GIL itself has "unlocked" the thread. The thread can sit around doing nothing until the OS either schedules a different thread, or until Python happens to unlock the same thread. The full story is pretty complicated ; look at http://blip.tv/file/2232410 . This is a classic case of the Law of Leaky Abstractions ( http://www.joelonsoftware.com/articles/LeakyAbstractions.html ). You also see this with network protocols a lot (layers and layers and layers of windows/segments/fragmentation modes/retry modes == incredible inefficiency).