(Due to keyframing, there's no clear image for the first few seconds of playback, nothing important during this time)

Description of problem, and what the video is trying to demonstrate:

Using Leadwerks Engine 2.27 with the solver set to the exact model, it seems to be possible to fly though walls if a body has a high enough velocity.

After failing to slam through the walls after two attempts, I set the physics to update slowly, and watch the collision bounds of the spectator body (green sphere) pass right through the oncoming wall. Basically, rather than checking of any objects are in the way, it simply teleports to its new location and checks for overlaps.

Indeed, it's also possible to build up velocity in slow updates, and then set the update speed back to 1 second per call whilst the body is in motion. If the speed is high enough when the update speed is changed, the body will still fly through walls (so, a high update timestep doesn't stop bodies teleporting through walls, it just makes it harder for the bodies to reach a sufficient velocity to behave this way).

The camera is positioned at the point of the spectator body every frame, so the fact that the camera position and sphere's visible position, indicate the that engine does indeed know where the spectator body is, even though the collision is being offset from this point.

I did not notice this behaviour with the collision boundaries when using the 1.32 Newton wrapper for DB Pro, with a timestep of 0.01. Although this was some years ago. (less info)