Right now, mochitest-other is comprised of 3 separate test suites run together for historical reasons: mochitest-a11y, mochitest-chrome, and mochitest-plugins. Once upon a time, mochitest-browser-chrome also ran as part of it before being liberated into its own suite. It would be nice if we could just run these suites individually for platforms that are being switched to Taskcluster and things like builder limits no longer exist. Is there a good reason for not doing so that I'm missing?

My memory of the origin of the Moth was not builder limits, but the ridiculously tiny size of the subsuites. Win7 opt, a11y takes 215s to actually run, plugins takes 64s. Dunno what kind of setup and teardown times desktop tests will have on TC, but the b2g tests (9 minute Gip job includes 3 minutes of running tests, 13 minute Gij job includes 54 seconds of running tests) don't give me the feeling that TC will make running a 64 second test suite a good efficient use of resources.

There are no reasons specifically to run those suites together- when it all started mochitests were run in a single job - the build job, then eventually mochitest-plain and mochitest-other, now a big list. :philor brings up a good point, and this is something I see in run-by-dir as well. Our overhead should not be 10 times that of actually running the tests. In fact once our overhead (total job time) is more than twice that of the actual test time, then in my opinion we have gone too far. on other platforms like linux, a11y does more than a few seconds- in this case of the bug we are starting with moving linux to task cluster.