Flaky threaded parallelism unit test

We saw `test_score_pipelines_passes_X_train_y_train[multiclass-cf_threaded]` flake on the linux nightlies on 2021-10-27.

Logs are [here](https://github.com/alteryx/evalml/runs/4018971157?check_suite_focus=true)

We've seen this flake a couple of times (@angela97lin saw it fail on her PR once) so we're recording it here to see how widespread the problem is.

Although this is similar to #2669, that issue tracks figuring out why the process-based parallelism tests (now removed) used to flake.

Closing out this issue involves figuring out why this test can flake and proposing a fix if possible.