The problem is that cheap lenses (and even some expensive ones) do not know where the focus ring is actually positioned. They only know if they are in focus and which way to run the motor to get more in focus. For focusing, they don't generally need access to this kind of information because they get feedback via the AF sensor about if they need to keep going.
When shooting video, the PDAF sensor is covered and the only way to verify positioning is to continue until contrast decreases. This is why you get the focus hunting issue with contrast based auto focus such as what is used on most LiveView modes (unless the camera has a hybrid sensor that includes PDAF pixels.)
What Magic Lantern is probably doing to get around this is running the motor for a fixed amount of time to approximate the movement, but in cheaper lenses, the motors are particularly imprecise, so it is going to often miss the mark. There might also be an option to have it use CBAF for confirmation, but that will result in at least a minimal amount of focus hunting.
There's still no real software fix for what is really a matter of needing to be able to move from position A to position B exactly unless you have a lens that actually provides exact position information back to the camera (in which case, this information may be able to be used by Magic Lantern for giving an exact follow or rack.)