Having designed and run preferential ranking systems for selecting multiple winners (unlike the Oscars), I've seen that polarizing choices tend to get eliminated regardless of artistic merit. Sometimes, we address this by letting the results of the larger voting body inform the decision of a smaller panel, who are allowed to vary the algorithm and play with other variables before making the final determination (in ranking subjective things like art, fairness in voting is not always the main priority).
The problem with the compromise candidate is that it is often not anyone's first preference, therefore nobody considers it the best. This is especially true when the number of candidates is large.
As an experiment, I'd like to ask voters to select only the best and worst single candidates, then apply a straightforward preferential ranking algorithm against those as if they were the top two choices. I wouldn't be surprised if this would narrow the field to works with the most artistic merit.