As far as I understand, cell-shaded cartoons are basically impossible to use motion interpolation with. The reason being that between any 2 frames, there's a multitude of possible "solutions".
I wrote that guide a long time ago, and I've actually noticed the problem myself even with the highest "Motion Vectors Grid" set. The only real use for SVP is for reducing the judder of zooms and pans.
It's been quite some time since I fiddled with it, but the trick lies in altering the settings behind the "Adaptive" interpolation mode. Setting the conditions for when to activate frame interpolation to the max makes SVP only activate under absolute ideal conditions (pans and zooms). I might aswell write it in stone here: Artifact-free motion interpolation for cell-shaded cartoons is impossible, even with the best algorithms.