I have found two similar situations where removing the sync handler from a pipeline's bus with set_sync_handler(None) results in a segfault.

Details / Reproducing:

I have two short test cases with backtraces.  In the first, I remove the sync handler after I have set the pipeline's state to null.
test case: http://pastebin.com/R7dtsj9J
backtrace: http://pastebin.com/rGZbAMXR
The segfault happens regularly on exactly the i==4123 iteration of the loop.

In the second, I remove the sync handler before setting the pipeline to NULL.
test case: http://pastebin.com/ByVQjqH5
backtrace: http://pastebin.com/XzfJeC92
This one segfaults with some probability (the loop is in the test only to increase probability of occurrence).  Also, in this case, I get a python error "TypeError: 'tuple' object is not callable" immediately before the segfault.

In both test cases, 'foo.wav' is some input file; different file sizes (incl. 128 samples and 60 minutes) didn't seem to affect occurrence.

System / Install:
Ubuntu 12.04
Python 2.73
python-gst 0.10.22-3
libgstreamer 0.10.36
libglib2 2.23.3
gstreamer0.10-plugins-base 0.10.36
(All gstreamer packages installed in last few months using apt-get.)

I also reproduced these segfaults on Debian Wheezy, with the same python, gstreamer, and glib versions as above.  On Wheezy, the first testcase (http://pastebin.com/R7dtsj9J) segfaulted on the i==4122 iteration rather than i==4123.