(I'm the author of that test.) Interesting! Now that I think about it, that test makes the assumption that the CPU availability will be roughly equal. Namely, it:
1) creates a file which should take 10 seconds to decrypt 2) attempt to decrypt that file in only 1 second. If the system was very heavily loaded while it did step 1 but the load disappeared before step 2, then that timing could well fail. Attempting to create the file during the tests was supposed to make it platform-independent: the test should take a reasonable amount of time, regardless of whether it's being run on a desktop or a raspberry pi 2 (which was literally what I checked when I wrote the test in 2016!). I'm looking into lowering the "deliberately too slow to decrypt" time from 1 second to 0.1 seconds; so there would be a 100x difference between the encryption and decryption.
