OK guys, since I cannot address each of your points on a personal basis because it would be time consuming, let's just get some things out of the way.
1. 8GB vs 4GB of RAM does not make sense due to multiplatform environment/dev costs.
Have you heard of "detail levels" as in low, medium, high detail? That's right, something that nearly ALL devs use RIGHT NOW for any game that is ported to the PC!
The fact that most of the developers do this
as we speak and manage to somehow take advantage of the fact that PCs have 4x to 8x as much RAM + VRAM without seemingly incurring significant extra costs should tell you something basic about game developement:
Developing a game always starts at the highest fidelity, which is then optimized or watered down to fit on particular platforms. Even if you do an exlusive, you'll want your source material as high fidelity as possible in the hopes that not all of it will be stripped away during optimization process and you will still have a looker.
So, in fact, nearly all AAA devs have those high fidelity textures and materials
in the first place. They are f*cking happy if they can use them on the final product, but the sad reality is that significant sacrifices have to be made for the final product.
More RAM equals less time spent on painstaking optimization equals less money burned equals a better looking game equals happier devs who choose your platform to be the lead platform and the rest get a stripped-down port.
And since one of the 3 lead platforms already does not have a RAM issue (PC), then it would make sense anyway because you can make TWO good-looking versions with minimal porting costs and then only one "optimized" version.
2. Stop with the Llano references, please. It simply does not apply. Llano is an example of the lowest end hardware applied on the lowest common denominator (4GB 1333MHz DDR3, can be bought at less that €20). Naturally there will be a bandwidth problem if the GPU is twice as powerful (regarding shader power) as that on the X360 while having access to *less* bandwidth. However, bandwidth will not be a problem for the next gen boxes.
Remember, we are speaking about 8GB GDDR5 here?
The fastest available GDDR5 right now was manufactured already in 2008 - it runs at 7GHz and one 32bit chip can put out 28GB/s.
On a 128bit bus that would be 112GB/s, on a 256bit bus it would be 224GB/s. Remember, this is
3 years old!
There is no reason to assume that by 2014 there won't be significant advances which double that bandwidth while providing 4-8x more density ... 8 of those chips made 1GB in 2008, and 8GB by 2014 is reasonable.
As for the GPU power to harness all that RAM? This winter we will have 4GB single-gpu cards with texture fill rates reaching 100 000+ Mtex/s (compare this to PS3-s 13 200 Mtex/s or Llano's 8 900 Mtex/s). By 2014, this number will be doubled, and then nearly doubled again. Think about that. By 2014 8GB VRAM will be as mainstream in the PC world as 1GB VRAM is now, and the GPU power to use that will be there.