This is a situation where varFDTD would not be the right tool to use. The reason is that the vertical profile of the waveguide changes drastically since the MQW layer is only present on one side of the structure. This can lead to vertical mode coupling as explained here:
Ultimately, you can try a full 3D-FDTD simulation to check your results.
Oscillations in the reflection:
I am not sure if the oscillations you see are an artifact of the simulation. They are more pronounced for the net reflection in the case with MQW, where varFDTD is definitely not appropriate. My suggestion would be to check if you see the same behavior using the EME solver. There are some additional things to check:
Distance between the waveguides and the simulation boundaries: As you increase this distance you might find that the simulation results change. It is a good idea to do a convergence testing for one angle to find the distance such that when you increase the distance further the results don’t change significantly. Once you find this, you can use this setting for all the angles in the sweep. Remember to make sure the structure is large enough so that the boundaries are inside the structure.
Metal or PML boundaries: The plots for total and TE reflected power show that some power is not reflected back in the fundamental TE mode. Probably, there are radiative losses at the interface between the waveguides. If that is the case, you should use PML boundaries everywhere.
Yes, you can simulate the effect of the tilted interface. Since the structure profile perpendicular to the propagation direction changes with position at the tilted interface you need to use several cells in that region. I suggest taking a look at this getting started example of a spot size converter. As shown there, you can approximate the continuously varying structure profile with a number of cells; the staircasing effect can be minimized using the CVCS method.