There are four written fiction categories whose boundaries are based on word count (40,000, 17,500, and 7,500 words are the boundaries). There are also two dramatic presentation categories, with 90 minutes as the boundary line. There is a 20% leeway in all cases except Novel/Novella, where the leeway is +/-5,000 words.
Technically, and by the narrowest reading of the WSFS Constitution, only the Hugo Administrator can relocate a work from its "proper" category to the adjacent one, assuming the work falls into the inter-category gray zone. I believe that the individual voters have the implicit right to make such relocation as well, and that they should nominate "gray zone" works in the category they think more appropriate for that work even if it appears to technically fall on the other side of the line.
Cloverfield is less than 90 minutes (but more than 72, the lower boundary of the gray zone), but it's obviously at theatrical motion picture. The definition of Best Dramatic Presentation Long Form starts out by saying "Any theatrical feature...." and that's meant, in my opinion, to suggest that theatrical motion pictures belong here even if they're slightly short of 90 minutes.
In a similar manner, a story that some word count methods show as 7,800 words may well be structured and identified as a short story, and if you are considering nominating it, you should place it in Short Story if that's where you think it best fits. If you think it's structured more like what you would consider intermediate-length fiction, by all means place it in Novelette. And if there is a short novel that is between 35K and 40K words published in a separate cover and marketed as a novel, it may well be appropriate to nominate it as a Novel, not a Novella.
People are concerned that the Administrator will eliminate their nominations if they nominate something in the "wrong" category. I'll admit that the narrowest reading of the constitution, particularly section 3.8.6, suggests that they will do so. But the other side of this is that Administrators are called upon to decide into which category a gray-zone work falls, and the place where more people nominate it is almost certainly going to influence their decision if they have any sense at all and have been paying any attention.
Most Administrators do pay attention to what's being said about the Hugo Awards, although they may not say a whole lot when they are the current Administrator. That's because they don't want to be seen at influencing the Awards. They'll rarely make hypothetical rulings, for the same reason a prospective US Supreme Court justice is unlikely to tell the Senate Judiciary Committee how s/he would rule on a hypothetical case. Furthermore, any reasoning about how an Administrator could deliberately take a perverse interpretation in order to play havoc with the system assumes that the Worldcon committee appointed people to their Hugo Administration Subcommittee who are prone to doing such things. If that's the case, your committee has a whole lot of problems, and a couple of oddball rulings are likely to be the least of your worries. (My thanks to Mark Olson for the wording from which I paraphrased some of this.)
There's a false assumption of sorts in deciding what the "right" and "wrong" category for a given work solely on length is. It assumes that there is only one single definitive answer to the question "How long is this work." With written fiction, there is no one way of counting words, and the common methods will produce a range of responses. Even dramatic presentations have some variability, although usually not quite as much as written fiction works. The variability is one of the reasons there is a gray zone in the first place.
It was pointed out on the SMOFS list that the 7,500 and 17,500-word boundaries are defined in such a way that a work that was exactly 7,500 or 17,500 words long would be eligible in both categories equally. The definitions use the word "between," which in my opinion (particularly influenced by my database design experience) includes the end points. I am not troubled by this "error" because it's irrelevant. No work can be that precisely measured, and anything where one of the counts hit exactly would have the other counts probably falling in a range surrounding that boundary point.
There's another side to this question, however, and that could be summed up as Don't Get Cute. Don't take advantage of the gray zone to nominate ten short stories (spread between Short Story and Novelette, with only works in the 6-9K range in Novelette) just because you can. You should honestly feel that the works you nominate in a category belong in that category. An Administrator who detects what s/he considers to be a deliberate "gaming" pattern is, in my opinion, likely to take a harder line on what to count than if the ballot only has one or two "offside" works. Furthermore, if you're making an "offside" nomination, it might not even hurt to include a note explaining that you know you're doing it, and why. Administrators are human, and human beings read all of the ballots, even those submitted electronically.
So there you have it. I think the voters have a huge role to play in deciding where gray-zone works belong, and that you should think carefully about what you are nominating and why, and not consider the rules as written to be as firm as they may appear to be on the surface.