in your example, the route is going over an area that is marked as a footway.
That means, that instead of following given ways, the routing must try to find the best path in an open space.
Open space routing is not as trivial as it might seems
For example, once there are additional features in an open area, routing would have to take these into account:
Imagine a bench on an open area. In OSM, that would most certainly be tagged as a point feature, but in reality it obviously has an extent that would block a potential path passing the point feature.
Then of course you have reality - think about large open spaces where bicycles are conveniently parked between trees, leaving just a passing here and there.
In fact, we have some open issues on this:
There has, however, not been a final decision on how the openrouteservice should handle this.
To me it does not look like the area issue, but the question is, why openrouteservice does not follow the line highway “Schießtäle” as other routers do. My example is not perfect, if you move the starting point a bit north the difference becomes more obvious. Or have a look at this situation:
your second example uses the wheelchair profile.
Switching back to foot-walking changes the situation again.
To be fair, I do not know why the openrouteservice does not prefer a routable way instead of a routable area.
My best guess is that highway=footway is, in a way, more permissive than a highway=service that has foot=designated on it - but this is only guess work.
There is a quite recent open issue regarding handling access restrictions:
I could imagine that these issues are related…
I will link this example both there and in the mentioned issues
