These are the two approaches I am considering:
1) Using standard axonometry (e.g. isonometry). In this case, I place meeples flat on the map. I think they look reasonably good. Vehicles or other items could be added using axonometric counters or tiles, but these tiles would be facing in specific directions. Meeples are great because they are quite generic and can fit in any position. Of course, what one gains in flexibility is lost in detail. If ones tries to use this kind of drawing with 3d objects, making the meeples stand or adding plastic soldiers or model vehicles, the result is a weird distortion of the buildings:
2) The other possibility I am considering is moving to perspective. Often in perspective vertical lines are drawn as parallel, while other lines converge to the focal point. In this case, it is necessary to have the vertical lines converge towards the camera, so that they result vertical in the picture. I decided to keep all horizontal lines unchanged: one can simply draw a 2d map with only the horizontal lines, and then add the vertical lines, making them converge appropriately.
In this case, 3d models fit reasonably well. But things are harder to draw and the distortion you get in “play view” (see the image on the left above) is unpleasant.