Currently, the threshold for approving or rejecting a suggested edit is two approve / reject votes, whichever is reached first.
This means that, when bad or just dubious edits get approved, it's often with a vote count of something like 1 reject / 2 approve (or 2 reject / 3 approve on SO). Typically, these edits manage to get one or two very quick approve votes from, um, fast and inattentive reviewers, after which it's very hard for the slower and more careful reviewers to accumulate enough votes to catch up before the edit is approved.
Conversely, when a valid edit gets wrongly rejected, this very rarely happens unanimously either. More often, you'll see something like 3 votes to reject / 2 to approve, but that single last vote, no matter how well or poorly considered, is enough to tip the scales.
So I'd like to propose a simple change to the approval / rejection threshold:
Suggested edits should be approved when S = (# of approve votes − # or reject votes) ≥ N, where N = 3 on SO and N = 2 on other sites.
Similarly, suggested edits should be rejected when S ≤ −N.
No, this change is not going to magically fix all the problems with suggested edit review. But I believe that it would be a simple and conservative improvement over the current behavior:
For obviously good or bad edits that get approved / rejected unanimously, this suggestion would not change the current behavior in any way.
For edits that get one mistaken or careless reject / approve vote (as happens pretty often), an extra approve / reject vote would be required to balance it.
This would slightly slow the review process for such edits, but not excessively so. Further, I would argue that these are precisely the kind of edits that deserve that little bit of extra attention to make sure the eventual decision will be correct.
Edits that are truly controversial, and for which the reviewer community is nearly exactly 50–50 split between voting to approve and reject, might stay in the review queue significantly longer (maybe twice as long on average) than under the current system. However, this is a trade-off for the increased likelihood that the eventual decision will actually reflect the majority opinion among reviewers, rather than mere random chance.
In any case, even if reviewers are exactly 50–50 split, the expected number of votes needed to reach S = ±N by random chance under the proposed scheme is only N², i.e. 4 votes for other sites and 9 votes for SO. Even a slight bias in the voting will further reduce this. Besides, the longer the suggestion remains in the queue, the more likely it gets that someone will short-circuit the process by casting a binding vote (mod / post owner) or by choosing to improve the edit.
Ps. If it is felt that there's an issue with approval being easier than rejection, a simple further chance to address that could be to make each reject vote count as, say, −2 rather than just −1 approval votes (and change the rejection threshold to S ≤ −2 × N to compensate).
This still wouldn't help with bad edits getting unanimously approved (which isn't really something that can be fixed, except maybe by slugster's cattle prod), but it would mean that each reject vote would now effectively cancel out two approve votes, titlting the balance for disputed edit suggestions in favor of rejection and making it easier for more careful reviewers to counter the "robo-approvers".