IMO test-with-buildbots should be the most trusted buildbots, like stable buildbots, maybe only Tier-1 and Tier-2. But not slow “Refleaks” buildbots. My concern here is to no annoy people with buildbots know to be unstable, but also not to waste resources. A single Refleak build can take 1 to 4 hours! Usually, Refleak builds are run on the same builders than other non-Refleak build, and so the buildbot queue can be blocked for hours because of that.
If some people want to run more buildbots, like running ref leak tests, test-with-refleak-buildbots should be used: but I would prefer that this common only runs Refleaks buildbots.
If some people needs specific buildbots, the !buildbot REGEX can be used.
If we need more buildbots, a new label can be added. The link above are about unstable buildbots like NoGIL.