Correct, I do believe.
In that case, the final object doesn’t have a scaling factor–it’s just a mesh that happens to have been made in part by scaling its vertices. The result isn’t significantly different to the case in which one moved each vertex to their final positions by hand.
The issue, as I understand, comes from having a physics/collision object that, in-engine, has a transformation that includes not just position and orientation, but also scale.