We want to run a tight ship in general, but a few of these tests are
intending one thread to block while the rest of the test does its thing;
the new 0.2s default seems to be a bit too short for reliable CI.