This is a common issue in simulating inductance of a single conductr: If you measure inductance, you need to think about the return path also. Current flows in closed loops only. What you measure is the total effect of both conductors. And to make it more complicated, there is coupling between signal and return path, so that distance between signal and return conductor (here: boundary) matters.
In your CST model, return current will flow through the boundary back from port 2 to port 1, and this physical path adds inductance. This current flows through the boundary condition, which is included but not shown in your model. With a small distance to the metal boundary, you distort the magnetic fields of the signal conductor. With a large distance to the metal boundary, the return path though the boundary is even longer -> more effect on total inductance.
It looks like a trivial simulation problem, but it is not!