You are right and using PML BC makes the most sense. However, simulation with PML BCs tends to run slowly and sometime PML creates non-physical modes. Please note that in cases where electromagnetic field decay enough before reaching boundaries, you can safely use metal BCs. This page discusses your questions in more details:
Regarding the second part of your question: This example tries to replicate the results of this paper (https://www.researchgate.net/publication/3409902_Microphotonics_devices_based_on_silicon_microfabrication_technology)
and below is the screenshot of the geometry taken from the paper:
Since the top layer is covered by another cladding, the background index of EME solver is assumed to be 1.465. Alternatively you could select background index to be 1 and add another object as clad and results will be identical.
Generally speaking, once you are modifying the geometry (in your case changing background index to 1), the internal s matrix, that shows coupling between different modes of input and output (for 10 modes on each port, this will be 20*20 matrix), will be different.
I also take a look at the coefficients (after clicking eme propagate, select EME and look at the result view) and compared it in two cases:
As you can see, when you change the background index to 1, most of the modes are coupled to second mode on the output (which is TE polarized). On output (port 2), the order of fundamental mode changes when you change the EME background index. To be more clear, when background index is 1.456, fundamental mode at output is TE polarized, while for background index of 1 the fundamental mode is TM polarized (while input mode is TE polarized in both cases). As a result, you are getting a poor transmission for background index of 1.
If you select fundamental TE mode on both port 1 and 2, you will get a better transmission when background index is 1 (more than 40%) because basically you will have a better mode confinement with air surrounding than cladding.
I hope this was helpful