My big python app (the robot end of a telepresence connection) suffers from GIL problems. Mostly, threads are used for C-only stuff like video compression so they can easily release the GIL while running. But I gotta say, the complexities of GIL correctness, thread performance, gtk/gobject, callbacks, boost.python, GC-correctness of both Python and C++ objects, and polling make my head hurt. Simplicity would be a find goal for the new system.
