The process variable 'p' is leaking into sub-processes when using the
multiprocessing modules.  The following code demonstrates the problem:

    import sys
    from multiprocessing import Process

    p = 'Correct'

    def test():
        print "Got 'p' of", p

    if __name__ == '__main__':
        if len(sys.argv) == 2 and sys.argv[1] == '-m':
            p = Process(target=test)
            p.start()
            p.join()
        else:
            test()

Running this in SP and MP mode shows the leakage:

    ssmith$ /opt/python-svn/bin/python mpbug.py 
    Got 'p' of Correct
    ssmith$ /opt/python-svn/bin/python mpbug.py -m
    Got 'p' of <Process(Process-1, started)>

This occurs in both 2.6b3 and trunk.

What do you mean, "overriden"? The behaviour looks totally normal to me.
You change the value of "p" before the subprocess is started, so the
subprocess inherits the new value, not the old one.

I agree; this is the expected behavior. Perhaps you don't understand how
globals work in Python?

Well, I get the OP's expected result on windows:

C:\dev\python\trunk>PCbuild\python_d t.py
Got 'p' of Correct
C:\dev\python\trunk>PCbuild\python_d t.py -m
Got 'p' of Correct

This is easy to explain: on Unix, the forked process has a copy of the
memory and reads the last value in the module. 
But on Windows, the freshly spawned process imports the module, and get
the initial value (since it does not enter the "__main__" block).

This is documented:
http://docs.python.org/dev/library/multiprocessing.html#windows , under
"Global Variables".

By the way, the Windows way may have some advantages for some uses.
Could the same method (start a new interpreter, import modules, copy
needed objects), be made available on Unix?

Ugh, sorry.  Stupidity on the reporter's part.