Try the following script on posix and Windows.  On Posix:

launched
. . . exiting
killed

on Windows:

launched
. . . exiting
Traceback (most recent call last):
  File "sp.py", line 16, in <module>
    p.terminate()
  File "c:\Python26\lib\subprocess.py", line 949, in terminate
    _subprocess.TerminateProcess(self._handle, 1)
WindowsError: [Error 5] Access is denied

This inconsistency seems unnecessary and is an obstacle to writing portable code.

from subprocess import *
import sys, time

p = Popen([sys.executable, '-c', '''
import time, sys
for i in range(3): 
    time.sleep(.3)
    print '.',
    sys.stdout.flush()
print 'exiting'
'''], stdout = sys.stdout, stderr = sys.stderr)

print 'launched'
time.sleep(2)

p.terminate()
print 'killed'

By the way, the suggested fix would be for terminate() to return a value indicating if the process were already terminated, and not throw an exception in that case.  For a user to handle the issue correctly on Windows is rather a nasty project involving a race between process death and the call to terminate()

Here is a patch for 3.3. I'm not sure it should be backported, as it's a slight change in behaviour.

Raising an exception on terminate is a bug.  I'd backport this to 2.7 and 3.2.  I don't actually have Windows to test on so i'll leave committing that to people who do.

+1 for backporting to the bug fix releases.

New changeset 41b1fe5a75a6 by Antoine Pitrou in branch '3.2':
Issue #14252: Fix subprocess.Popen.terminate() to not raise an error under Windows when the child process has already exited.
http://hg.python.org/cpython/rev/41b1fe5a75a6

New changeset f452d7d5470d by Antoine Pitrou in branch 'default':
Issue #14252: Fix subprocess.Popen.terminate() to not raise an error under Windows when the child process has already exited.
http://hg.python.org/cpython/rev/f452d7d5470d

New changeset b6ec3b717f7e by Antoine Pitrou in branch '2.7':
Issue #14252: Fix subprocess.Popen.terminate() to not raise an error under Windows when the child process has already exited.
http://hg.python.org/cpython/rev/b6ec3b717f7e

Ok, I've then fixed the bug in all 3 branches.