When xharness aborts a test run on android due to hitting the --timeout, it kills the adb shell am instrument process, but this doesn't cause the instrumentation process on the device to stop.
I don't know how the helix infrastructure handles this (whether it restarts the emulator per run, etc), but it seems like this could cause unreliability in our test infrastructure.
We should consider handling timeouts by also issuing an adb shell am force-stop <target-package>.