I wonder, if the assumption you made, is correct in all circumstances. I think it was Eric Bogatin, who mentioned in one video, that you can use a power plane as the reference plane, if it has a low impedance connection to the driver of the signal. That make sense to me, since the lowest impedance path for the displacement current would be right through the driver's power pins then. Intuitively I would expect, that you could get a little more noise on that power rail, but the current loop would still be well defined. When the referenced power plane does not meet this requirement, you are absolutely right. The displacement currents can go anywhere. From the overall design perspective, I would only use that as a last resort though, since you have even more things to watch out for (split planes, coupling to adjacent planes etc.).

Let me just give you my 2cents on what you laid out so far: Regarding the 2-layer-board: Probably not the best decision to go for only 2 layers. 1. You are using switching regulators, that usually create high frequency ripple currents that in turn cause EM radiation if not confined in a proper way. Always think of where current flows and try to minimize the area of that current loop. Furthermore make sure that fast switching signals see a constant impedance, which usually means having a reference (ground)-plane next to the signal layer. 2. I would go for at least 4 layers with a solid ground plane underneath the regulators. That helps to confine the EM-fields to a small area and provides a proper return path for the signals. The ground plane will also help to distribute the heat away from the regulators, especially when you can connect it with lots of vias to the large ground pad (which the regulator probably have). If you absolutely have to stick to 2 layers, you could try define shield rings around the noisy nodes. But that will be much less effective than a continuous reference plane. Heat distribution is also more difficult if you only have two layers. It could work if you can do the routing (mostly) on one layer and use the other layer as a ground plane. Additionally you could use 70µm copper layers to increase thermal conductivity but I would rather go for 4 layers due to the EMC benefits. 3. For the 6-layer board it should not be difficult to keep temperature in check. Again, connect the ground pads of the heat producing components to the ground planes. 4. Regarding EMC I would make the stackup symmetrical, with the 5th layer being a ground plane. You want to confine the EM-fields of your signals. So always keep in mind, what your signals couple with via capacitive and inductive coupling. Try to avoid coupling to other signals and split planes since that will increase the risk of EMC problems. 5. If would distribute power on layer 3 and/or 4. Ask yourself if you really need ground planes. Often you easily get away with routed power. Then you can probably make layer 3 or 4 a solid ground plane too which further benefits your EMC performance. As far as I can remember I never really needed more the 3 signal/power layers when no BGAs were involved. It is all a question of proper component placement. 6. Watch videos with e.g. Eric Bogatin and Rick Hartley. These will help you a lot to understand the core principles.