Gimp abstracts differrent frames of animated gifs as layers. You may like it or not, (as well as the inability to link layers) but not call it "incorrect".
Because GIMP is designed to work with images of all sizes, those computational costs must be considered for operations like rotation. The "preview" of a rotation needs to occur very quickly, because a user might be rotating an image by hand using a rotation tool. You need redraw speeds that are >30 fps in order for the rotation to appear smooth at all. When you commit to the rotation, GIMP can use a more computationally expensive algorithm. This algorithm will deliver different results when observing at a pixel level.
There are still open questions here, like why is the base layer altered at all? But I suspect that the reasons I've outlined above.