While working on BF-30427, I identified and fixed a true data race, and filed tickets for the others. 

However, I it was hard to reliably test the noPassthrough task without hitting the timeout, and I raised the timeout to 3 hours for the duration of my testing, which I found allowed the tests to reliably complete under the time limit. I think it may be worthwhile to permanently increase this time limit in order to reduce the flakiness of the TSAN testing.