But this could conceivably cause more�� erroneous failures by the CI
monkey, which slows down progress up to the Release.

+1 for doing it after the Release.

Now what would be interesting (though maybe not feasible) would be
recording how long each test takes, maybe sending to a central server,
to do maybe do statistics on them.�� Maybe faster running tests could
be scheduled first.�� Or we may find some tests have multiple modes to
optimise.�� Or which tests fail the timeout more often - to help set
the timeout.�� Perhaps the timeout needs to scale for different
environments - like running of a RPi rather than a beast desktop.