In multiprocessing, a pool worker process would die
if the result/error could not be pickled.  This could
cause pool methods to hang.

In 3.x this was fixed by 0aa8af79359d (which also added
an error_callback argument to some methods), but the fix
was not back ported.