If the earth were truly inhomogeneous in all three dimensions, we could hardly expect the data of a single seismic line to make any sense at all. But reflection seismology usually seems to work, even when it is restricted to a single line. This indicates that the layered model of the earth is a reasonable starting point. Thus normal-moveout correction is usually a good starting process. Mathematically, NMO is an excellent tool for dealing with depth variation in velocity, but its utility drops in the presence of steep dip or a wide dip spectrum.
My early migration programs were based on concepts derived from single profiles. The data and the wave equation were transformed to a moveout-corrected coordinate system. This approach to migration is well suited to data that is sparsely sampled on the geophone axis. When steepness of dip becomes the ground on which migration is evaluated, then moveout correction offers little advantage; indeed, it introduces unneeded complexity. Whatever its merits or drawbacks, NMO commands our attention by its nearly universal use in the industrial world.