The VLM author’s intention is to have VLM’s output match that of the drone. You have shown, that in the case of the Mini 2, it obviously doesn’t. When time allows, I want to try a similar mission to yours to see how each of my drones behave. I fly a lot of waypoint missions. I just haven’t tried one similar to yours. To be fair, VLM’s output does match that of the drone in most cases. Yours is what I would call an “edge” or “extreme” case. Normally, that quick twisting motion is not what is desired in a video, unless that is specifically what you want to show.
If you wanted to pursue this, you could describe what you have found in that VLM thread that I linked. Namirda does watch that thread and normally responds when there are questions about his software.