I recently updated the buildbot host to the latest 9-STABLE sources (changeset delta is approximately 4-6 weeks)

On a hunch I asked our src committers if there had been any sendfile related changes and was pointed to 3 changesets that were merged from current (MFC) to stable/9 4 weeks ago:

http://svnweb.freebsd.org/base?view=revision&revision=250907

I reverted the kernel to r250906 (hunch commit -1) that resulted in a clean run:

http://buildbot.python.org/all/builders/AMD64%20FreeBSD%209.0%20dtrace%203.x/builds/1563

Manual Rebuild: Trying to isolate test_trailing with older kernel

I have it on preliminary evidence that our regression tests are passing on this functionality, but analysis at the Python end for a root cause and details on how the test implementation plays a role would be helpful.