It appears that the power-efficiency for second-order nonlinear optical processes is proportional to the square of the nonlinear interaction length along the propagation direction (~ longitudinal_distance^2) and inversely proportional to the nonlinear overlap area (~ transverse_distance^-1) if the pump power is kept the same when changing these sizes [http://www.opticsinfobase.org/josab/abstract.cfm?uri=josab-29-8-2199].
Is there any intuitive explanation for such scaling, especially taking into account that it applies to both spontaneous and stimulated processes?