To clarify: ideally (and likely I'm just stating the obvious) the thresholds we set for acceptance reflect the stages of the development cycle.
At the point we're currently at in the cycle - when the CVS HEAD has just been opened up after a new release - we should be working to clear away the backlog. This is the ideal time to be applying new patches: when we still have the rest of the cycle to work with. This means we have, probably, months to iron out any small remaining details, and also to take advantage of possibilities opened up by the changes. Or, if major unforeseen problems show up, to roll the change back--it wouldn't be the first time, and it's not the end of the world. If there's a time when we have maximum tolerance for small issues or instabilities in CVS, this would be it.
As the cycle progresses, we have less time to work with, more need to be ensuring stability, and hence diminishing ability to apply large changes.
In sum: early in the cycle - now! - is the time to be clearing the deck, applying any patch that has has obvious support, has addressed any major issues (leaving aside minor details or small remaining discussions) and applies without raising errors.
Who said this differently? It is just a matter of defining what constitutes "minor details". Goba