simply different thermal coeffs of expansion
If the potting has strong adhesion a thick rigid layer of it just wins, the PCB will have to expand/contract in thickness cause it isn't going to bend the potting. The tail does not wag the dog.
Yes, and imagine the forces involved in making the resin "win". FR4 can take this no problem. Now apply this force to the MLCC body. Board did not flex, but MLCC failed anyway. This is why MLCC-compatible potting compounds are not that hard, they need to be somewhat flexible, reducing the forces. But the problem is here: MLCC bodies are surprisingly sensitive, even small forces from soft compounds are risky. This is why this shit is actually designed and evaluated by very experienced people.
I'm no expert on choosing the potting compounds, but I know enough to know it is far from trivial and have read stories of smaller companies wasting years with poor initial assumptions. Personally I just sidestep whenever possible, mounting heavy components using silicone or other types of glue, instructing sealed enclosures, possibly with a silica gel bag strapped inside.
I have also read about using a very soft intermediate layer just on the top of shear sensitive components (MLCCs), around which hard potting compound can be used. For small MLCC sizes, bog standard conformal coating alone might be enough, but small packages are lesser of a problem anyway.
I talked with some potting compound experts, who actually encountered with the problem in the past. Stuff failing because a metal enclosure was used as the potting base, 100% filled. Then the thermal expansion was directed into the board. That's not how these are made, there is a separate block of potted electronics that is then fitted into the enclosure. If you make the PCB thicker, it will bend less. Similarly, if you pot it in a few cm of resin or urethane, it will bend less, and cracking will be lesser of a problem.
So I'm saying this, because it coincides with the info I get from the experts.
Now I totally agree with this. Your story made full 180 degrees from: potting is used to reduce cracking, to: with correct potting, cracking [caused by potting] can be "lesser of a problem", exactly my original point.