Gamma correction currently uses several pow instructions to convert diffuse textures to linear space. Each pow instruction can extend shader length by two or more instructions, so it would be a good idea to start moving everything over to 2D LUTs to minimize this cost. Could also look into applying the conversion LUT to the diffuse G-Buffer in the deferred renderer to reduce the cost of applying it to all partially visible objects in the scene.
Alternatively, we could just move over to approximating a gamma of 2.0 by multiplying the diffuse texture against its self, or finding a way to approximate a gamma of 2.2 without the use of pow().
Update: Instead, we're going for approximating pow(..., 2.2); and pow(..., 1.0/2.2); using a polynomial function that costs less taxing instructions than a single pow.