Each patch takes ~12h to pass today,
and while this is already a long time, it would be nice to enable even
more tests for each patch. To make this reasonable, patches should test
in several stages, and only continue if they don't fail too many tests.
Ideally, we could restart the stages separately, to avoid huge delays in
case of some small failure. Posting the results to Maloo incrementally
(i.e. after every test script completes) would also be useful. For this
to be useful it would be great to avoid repeated node provisioning for
each test.