As noted in bug 1527120 comment 0, we have some known regressions that are caused by Dynamic Reflow Roots -- and bug 1547305 seems to be a newly-discovered regression that it causes.
In bug 1527120, I did a "spot fix" for Firefox 66 to pref off the feature, with the hope of addressing the regressions before 67. But that didn't happen -- and until we've got time to address these regressions, we should guard Dynamic Reflow Roots from making it to release.
I think a "don't ship past early-beta" guard should be appropriate -- we can keep it enabled in "early builds", since the regressions aren't catastrophic and the feature is only activated in rare specific cases (I think). And there's value to keeping it enabled for these early builds, since it helps us keep it working & discover potential unknown breakage.
(Also: fortunately, this isn't a feature that's visible to users or web developers, really, so we can disable it without anyone really noticing.)