I would imagine a big advantage for the production team is being able to screen it and pause on any single frame and have it be perfect - consumer codecs all seem to use keyframes, which certainly save a lot of space, but if you start on a non-keyframe you either get no output or pixel puke until the next one.
They mark which frames they're referring to, so you at most just get a little stutter to decode those other frames if you're optimizing for the scrubber use case.