I initially noticed this when E froze up after clicking on a gadget (the backlight gadget). After attaching GDB I observed that both the threadq submitter and background worker thread for evas async sample sampling were stuck waiting on the threadq semaphore.
The ecore test suite can hit this as well:
$ ( cd src && tests/ecore/ecore_suite Ecore_Test_Ccore_Thread_Eina_Thread_Queue ) Running suite(s): Ecore msgs done 10000000 messages sent ERROR ERROR: DEADLOCK on lock 0x80a011008
(The deadlock detection isn't perfect — it seems to also hit this without printing a warning.)