Imho there are good reasons to use buffered data to some extent, especially for such a contest. I agree that it defeats the purpose if we're allowed to store 10 meg models inside the source, but for small stuff which cannot be created on the fly this is a great feature.
