I'd enable bullet-mode for the red body and leave bullet-mode off for the platform body.
At least in the C++ CCD code of Box2D 2.3.2 (see b2World::SolveTOI
), all contacts for the given time-step will be reviewed for whether CCD processing needs to be done for them. Essentially contacts having a bullet enabled body or having a non-dynamic body (a static or kinematic body) continue in the CCD process. So the less bullet enabled bodies, the sooner that method can finish.
As such, it's less optimal (algorithmically-speaking) to enable bullet mode on both the red body and the platform body though the difference in performance is unlikely to be noticeable with just the 3 drawn bodies. And given that you're only concerned about collision between the red body and the platform, one being bullet enabled is enough to get their contact to continue on for CCD processing.
As for why I'd bullet enable the red one (as opposed to the platform), that's because, as you state, it'll be moving the fastest of the two and that matches better in my mind with the bullet concept.
Note that if - say in the future - you wind up having more platforms and the platforms themselves need to be unable to tunnel through each other but you were noticing them tunnel (even at their slower speed), then bullet enabling the platforms (instead of the red body) would be the way I'd go.
bullet
? – Marco13