Let me reply to my own question since people have asked me whether I've been able to find a solution.
Basically, from my experience, GameCenter in Sandbox has serious issues. GameCenter in production has fewer issues but it still has significant issues.
For example, Axl & Tuna uses GKScore.context value in its scoring system. Well, in production, for about a week those values mysteriously disappeared. After I raised an issue with Apple, they eventually re-appeared. I've also witnessed the app successfully submit scores that never made their way into the leaderboards. Joy!
As far as getting duplicate scores back from GameCenter, or getting scores returned in the wrong order or with the wrong rank (both of which I have witnessed in production), what I ended up doing is scrubbing all the data returned from Game Center manually - ie. ordering it correctly, calculating rank correctly, and removing duplicate values when returned for the same user (and in my case just keeping the highest score, since that's what I wanted).
It's not that GameCenter is broken all the time or in a reproducible way. However, it's broken enough that I found score scrubbing a necessary work item to account for the 1% of cases when it failed, which is too often to just ignore.