Then I'll go for a "reconciliation post" as answer, because it's probably least confusing in the unlikely case anyone happens to stumble across this 🙂
The workflow that works for me now:
I am creating auxiliary datum planes in each component that correspond to the positions of the "optically relevant" planes. These planes are the ones that would have to coincide if the microscope was build straight, and the distances of these planes will also be used to lock all of the relationships.
The folding mirrors will realistically be placed under an angle of 45 degrees only, so I'm creating auxiliary planes as well that correspond to the planes that the light path is located in before and after the reflection off the mirror. Since the mirror is at 45 degree, these planes will be at 90 degree to each other (and 45 degree to the mirror).
The distance between the relevant optical plane of the component just prior to the folding mirror, and the relevant plane in the mirror is then defined when setting the constraints during placement. The distance between mirror and the component behind the mirror has to be the negative of the distance in the first constraint, and so on, through the whole assembly.
The positive and negative distances are linked by defining
$dist_neg = - dist_pos
in the relations window.
It of course helps to rename the distances first: This can be done eg. by editing the feature (right-click in the feature tree and select the icon with the distance pictogram), then selecting the distance (it has to turn from yellow to red) and then right clicking to select "Properties".