Most of the graphical features that people really care about were introduced on the PC in the late 90s and in consoles in the previous generation ... These are things like full facial animation and enough geometry and texture data to produce a representation of an object or environment that allows people to instantly recognise what the object represents. So, you could argue that we have already passed a graphical sweetspot.
Another way to look at this is when do we hit a point in processing power when the improvements made over an entire generation were fairly minimal and difficult to notice ... While I would say that we have already passed that point, I do believe the next generation of consoles will produce graphics where the main improvements will be from advanced lighting provided by global illumination algorithms.
