I think the idea of getting an ordering by pairwise comparisons (instead of discrete scores for each item) is sound, but it doesn't work as well for movies as for people.
pickthehottie.com (launched at least before June 22, 2001, according to the Wayback Machine) did the same thing to AmIHotOrNot's photo-rating juggernaut (launched in October 2000, according to Wikipedia). I think the A-B model worked really well for them, and the incentive for sticking around to compare for photos was a little more intrinsic.
How about having an option "can't compare"? That the user can indicate they saw both but cannot compare. With enough data, maybe you can cluster more frequently compared movies together or even making categories automatically?
(Downside is more complex UI; it's better to keep UI simple, so I'm not sure).
I guess I'm saying I would like it better if you had like different genres, they don't even have to be that specific like say horror action comedy family; and could compare withing those, but I guess that is limited with your selection right now.