I've found that the density Laplacian is not computed correctly inside xc_rhogen.F; furthermore, the AO hessian is not updated if pre-screening of grid points is performed in grid_pscreen.F. Perhaps this is the reason for the XDM parameters to be different in NWChem.