It's not a difficulty prediction tool, it's an endgame estimation tool.
If we know the $/gh, w/gh, and $/kwh we can estimate how much hashrate it would take before nobody could afford to continue adding hashing power. (but not when or at what rate)
We don't know the variable so it's only a guess but his point is clear that J/GH does make a huge difference.
So the graphs are done assuming a static difficulty for 2 years. Then of course it looks like you can make a profit, duh. When I say it is difficult to even recover capital costs of hardware, what I mean is given that you buy the hardware at a specific point in time, and then it mines in an environment where difficulty continues to rise at the historical rate. You know, reality.