For me, the novelty of a code golf site is seeing the novel ways people apply the features of languages I am interested in. As a result of this, it is quite likely I will be upvoting answers which are interesting to me, rather than worrying about testing shorter solutions in languages in I might not be interested in (I am not only referring to GolfScript here.) An accepted answer is only worth 1.5 upvotes, so if there are a lot of users like myself on the site, such solutions will consistently underperform in terms of rep gains, even if so called popularity contest questions are made CW. There's no point penalizing anyone for answering what some might consider an unclear question. All puzzles are objective questions to a point. We either need to decide that those questions with less strict acceptance criteria are off topic or accept them for what they are.
I suggest that it would be sufficient to emphasize the importance of appropriately setting expectations for acceptance criteria in the faq, and perhaps require the use of a set of (less condescending) acceptance criteria tags eg: strict, popular-vote, or whatever-i-feel-like
The ability for the community to upvote and downvote questions is a feature of the StackExchange platform, not a bug. Let's not try making it into something it's not. There's nothing stopping anyone from making a great code golf site with automated testing/automatic assigning of points to the shortest answer. This isn't that site.