The current run_tests.sh doesn't aggregate the test results properly - the results for the core tests, CLI tests and Web UI tests are all reported separately. run_tests.sh should be replaced with a run_tests.py that correctly collects *all* the test results into one count of passed and failed tests, as well as more cleanly skipping the CLI tests if the local Pulp server is not running.
0.0.16 fixes the test discovery problem, along with a problem where some tests were leaving temporary directories lying around and rsync daemon processes active.