The King of the Hill tag could use a "Suggestions" section to help posters avoid pitfalls that are KotH specific. It could also use a tip on what does NOT constitute a KotH challenge, since I have seen newbies using it incorrectly.
Note that just because all submissions are competing against each other does not make it a King of the Hill challenge. The submissions must interact. For example, a challenge where are submissions are all competing for smallest code size would be a code-golf challenge, not King of the Hill.
Suggestions for a Smoother Competition
Unless you want participants to abuse a particular aspect of the challenge (which is often the case), consider using the following restrictions to close loopholes that are specific to King of the Hill challenges. See <insert help page here> for suggestions that apply to all challenges.
- Allow one submission per user.
- Restrict access to other submissions and any data they may generate.
- Restrict modifications to standard libraries.
If multiple submissions come from a single user, then the challenge can degrade to "Who can submit the most answers". It is in the user's best interest to write many submissions that help him achieve his goal of winning.
Submissions can cripple other submissions if given access to their code, their data, or the standard library. You might argue that other submissions could react, but then it really boils down to whose code gets run first. If I go first, then I win by stealing your strategy. If changing the order of the submissions also changes the results, then your spec has a serious problem.
Another argument might be, "A perfect controller program would enforce these restrictions in code." The counterargument is that a perfect controller can only be written by a perfect programmer. The point of King of the Hill is not to pit the author against the masses.
Things I don't know how to address:
I do not know how to objectively stop multiple users from collaborating, or one user with multiple accounts. Does anyone have a suggestion that would allow someone to objectively determine whether a submissions was collaborating? Is this even a fixable problem? Let's see what the Game Development people think.