The new NoBodyPartThree.java test fails often when the fix is not present, but not always. If the server manages to close the connection before the client tries to reuse it, no error will be observed.

It occurred to me afterwards that it should be relatively easy to verify the fix with a small ServerSocket based test, that could fail if the server receives more than one last chunk (0:CR:LF:CR:LF).
Such a test would fail 100% of the time if the fix is not present.