Found it! It's caused by the additional call to
world.buildAllObjects();
in the demo Activity. This call resets the rotation pivots to the calculated values where they should actually be (0,0,0). The solution is to simply comment out that line and all is well. This should happen in desktop jPCT as well. I haven't checked, but I assume that the desktop example simply doesn't call this method!?
Why is this different in 1.24 and prior? Because of this change in 1.25:
Fixed behaviour of build() when used on models on which calcNormals() has been called before.
Before this fix, build() does actually nothing when calcNormals() has been called before. After the fix, it does it's usual calculations except for the normals and that resets the rotation pivot as well.
Just remove that line mentioned above and you should be good to go.
Edit: I should mention that calcNormals() is part of build() for a better understanding.