Assume that it is possible to prepare a macroscopic system (say a 1kg iron sphere) in a superposition of two position eigenstates 1 meter apart. This experiment has to be isolated from the environment, so let's assume that we place it inside a magical box that does not let any particle disturb the system. Outside the box we place a torsion balance. As far as I can tell the following premises are true:
No matter how the "magical box" is built the gravitational field of the iron sphere will still be detectable by the torsion balance.
The torsion balance actualy measures the object's position because the gravitational field points to the object's center of mass.
Once the position of the iron sphere is known, its state "collapses" into one of the two eigenstates.
Any massive object can play the role of the torsion balance.
From 1,2,3 and 4 it follows that such a superposition state is incompatible with the presence of any massive object outside the box. But in order for this experiment to make sense you need at least one object outside the box, the observer. So, it appears to me that such superpositions cannot exist.
Note: it is always possible to have a superposition as long as the measurement error implied by the uncertainty principle is enough to "protect" it. So the above argument is compatible with experiments such as the two-slit experiment. For the proposed experiment with the iron sphere the uncertainty principle is not relevant.