Yes edram is always active but I am talking here about the comparison of an Iris Pro CPU running at 55W vs an HD 4600 one running at 47W. Edram apparently even active needs barely 4W and it does give a bit of a speed boost to lowering the latency for quite many memory accesses which would otherwise go into main. Intel says some 95% hitrate in the first 32MB alone. No idea what use case but I assume almost always it will pay for its power consumption.
There is a few Watts room to waste on clock speeds. The GPU can be almost entirely and also partially power gated. I doubt it needs any more power than a GT2 when not really in use.
Sure if HD 4600 Haswell already stay at that high a Turbo given the slightly lower max Quad Turbo there isn't much room. I suppose while one can change the TDP the OEMs won't be allowed to add Turbo bins or else the 4950 vs 4850 difference would be ridiculous and the higher price kind of a waste. But maybe it is just the marketing. The spec sheet cannot say 4950 if it isn't.
Just checked on some info about this adjustable TDP and only found something on wikipedia.
http://en.wikipedia.org/wiki/Haswell_(microarchitecture)
I didn't find any source for that but those cracks who edit wikipedia in the tech sections are quite often right because otherwise they leave it blank.
And a 3.3Ghz base clock is something.
And if the the max short term Turbo power ceiling actually can go to 69W the GPU alone probably cannot do that. There ought to be more Turbo bins above the 3.3Ghz.
Given that we only would see and tinkering with a TDP if they remove the dedicated GPU while sticking to the rest of the design of the notebook, that would be a bit more speed for all day tasks and everyone who isn't a gamer. For GPGPU stuff Kepler sucks too. It is really only for gaming a great GPU.
