I wonder if there is a calculation error in the paper. EDIT: No error!
We may use particles with a rest mass m in the experiment. We can implement a beam splitter with a wall with a hole in it.
Let us increase the mass m enough, so that the system behaves like a classical system. Then we may assume that the particle did take a definite path through the system. If it ends up at D_2, it definitely did not kick the double-sided mirror M to the left, contrary to what the paper claims. It either did not kick at all, or kicked to the right.
The paper talks about a photon "taking a path" through the system. We do not use such language in standard quantum mechanics. The paper is suspicious, even though it was accepted to a peer-reviewed journal.
If we analyze the classical waves, if we set r = t, then M gets no net force and no photons reach D_2. If we tune r > t, then some photons reach D_2 and M feels a force to the left. Classically, it is not that the flux to D_2 "causes" the net force on M. The tuning causes these two effects.
The authors claim that if we shoot, say, 1 billion photons, and by chance they all end up in D_1, then we can measure no impulse on M to the left. Is that really true?
EDIT: Yes! One billion photons cannot be wrong. The system probably behaves in the experiment like a classical system which produces the same result.
Let us take a simple example. We shoot 1 billion photons through a beam splitter which is just a mirror with a small hole in it. If it happens that all photons pass through without changing their path, then the impulse on the beam splitter is zero. It is like having a beam splitter with a 100% transmission.
Analogously, we may guess that the Aharonov et al. device would behave like classical system where transmission is 50%. Then there is no impulse on M.
Note that if we have a classical wave in the Aharonov device, and the transmission is < 50%, then the intensity at D_2 cannot be zero "by chance". The intensity is always > 0 for a classical wave.
The Aharonov thought experiment shows that, if we by chance, measure the system behaving like a system where transmission is 50%, then the result is like for a classical system where the transmission is 50%. The classical limit imitates what we actually, by chance, measured.