I believe that the direction of Delbrueck scattering should be the
same as the direction of deflection of any uncharged, polarizable
object by an inhomogeneous field. As a photon passes a nucleus,
it "splits" into an electron-positron pair. The electron is
attracted and the positron repelled, so they acquire a dipole
moment pointing away from the nucleus. This dipole is then
attracted by the inhomogeneous field of the nucleus.
I'm a bit uncomfortable with such a semi-classical argument, but
in this case I'm pretty sure it's correct.