Just trying to figure it out myself.
There's 32 MB of ESRAM.
1600 * 900 (resolution) * 2 (double buffer) * 4 (RGBA @ 32 bit) = 11,520,000 bytes.
That's 10.986 MB out of 32 MB.
Deffered Lighting (and Deffered Shading) require additional buffers of varying bit depths in order to do their thing. They are becoming extremely common methods for rendering modern AAA games.
With these methods, there are so many variants of the buffer layout that its difficult to give a baseline estimate for the amount of memory and bandwidth needed for all games. The kinds of buffers and the way in which you will represent their data is all decided and prototyped out when you're still deciding what kind of materials and visual phenomena you want to be able to recreate and to what degree of accuracy. Some games want to have anisotropic specular, other want shaders for SSS, pearlescence, or some might need to model other complex visual phenomena... Some of these effects may end up needing extra buffers (although for some effects there are sometimes clever ways of getting approximations without additional buffers) or bandwidth. Some buffers need high bit depths in order to avoid artifacts like banding.
For some layouts, there can also be a handful of clever tricks that take better advantage of all the available ESRAM and its bandwidth due to some hardware level features of AMD's GCN architecture..
The point is its complicated to balance all the features you need for your renderer with the ESRAM and it's bandwidth, but I think the consensus is that, as time goes on and developers have time to figure out the optimal buffer layouts and approximation techniques, Xbox one games will reach the holy grail of 1080p 60 fps visuals. Its also not always easy to iterate on the layout late in development, especially in the case of multiplatform games.