There really isn't much disadvantage as half precision is still plenty for color blending, but GeForce FX users are probably the only ones that'll see much improvement since Radeons can handle floats in PSs much better, PS 1.x compile targets will likely compile down to using fixed point, and PS 3.0 targets can handle the floats without a problem.
I'll probably just leave it at half precision, as there is no discernable quality loss, and it will allow GeForce FX cards to run much better. Depending on how things go I might need more precision in the future though...(you can't do good per-pixel lighting, normal mapping, and parallax occlusion mapping in half precision)