what's "a.russell's method"? it was just an example i think. you can't use the two models like that.

what you theoretically could do is putting a shadow map into the alpha channel of a duplicated model and let it render over everything. do you mean something like that? it wouldn't really be a light map though but only a shadow map. i tried this once and it was fast enough even with a 50000 poly world. i think the reason it is fast enough is that the z-buffer is already filled after rendering the non-transparent geometry and then only the visible shadow geometry pixels have to be drawn.