yes it looks good…
just curious, why not to test the math with a simpler and clearer normal-map ? i mean, with your sample you cannot really tell if the math is right or just the “look”. it may just look “right” cos’ the esthetics look “right” 
first found on google :
http://img849.imageshack.us/img849/1398/1squareoutnormals.jpg
this should stick out.
(watch out for Y-flipped maps if you search teh interweb)
The best thing to test these will be the suzanne model from Blender. It’ll be perfect for testing lighting.
not yet :-\
the normal-map describes something that “sticks out” … so when the lightsource is on right, the left edge should be darker then the center and the right edge should be brighter.
the side that points towards the lightsource looks correct but not the opposite.
does it make sense to you ? 
I’m sure the normal map works. I think the thing wrong (if anything) is probably the specular normal.
[icode]- +specular;[/icode]
edit:
Removing the mouse light:
last image looks good … what happens if you try general gray instead of black ?
Turns out, my mouse-light was at the Z position 1. So that may have caused the problem :persecutioncomplex:
that looks nice 
