If the bandwidth is higher than than the data demand, your assertion is not logical IMO. 10 or 100 people can walk through a gate at the same time if it was designed for 500 people to walk through!
You're really grasping on this one, it's not about bandwidth here. It's about workload. Just because the same GPU is driving more pixels with one display than another doesn't mean it can do so with the same speed.
The bandwidth of both the discrete and integrated chips are significantly higher than the highest required resolution (3840x2400). I calculated it and the values are in a thread on this forum somewhere...
It's not about the bandwidth.
Several people have already tested using an external monitor, claiming that the rMBP is less "laggy" on the external, even when logically it is driving more pixels overall!
Actually, logically, when there's no redrawing going on for a given display the GPU isn't "driving" anything. Hooking a 10,000lb trailer up to a truck but not moving it doesn't put any stress on the engine. The external display isn't a mirror of the first, meaning the GPU isn't logically seeing the external as physical extension of the first and as a result sending all graphics I/O to both displays.
The GPU will perform the same with an external display whether the lid is open or closed, what will affect performance on the external is activity on the internal display because THEN the GPU is having to do processing for both displays.
"Lagginess" is not equivalent to UI performance, because the bottlenecks are not being limited by the UI, but by other parts of a program. For example, the problems Chrome had (and probably Safari) had to do with bugs in how images were decoded by the cache mechanism. This caused a perceptual "lag" but it was not rate limited by any graphical drawing or pixel throughputs, but by suboptimal background code that had nothing to do with copying pixels to the GPU buffer... 😉
That's all beside the point, IMO. The UI performance test is doing the same test on two different displays. It's the same GPU and the same software.
What you're basically saying is that a game that runs at 60fps @ 1440x900 should run at 60fps @ 2560x1440 because the GPU has the "bandwidth" to drive all those pixels at the same time.
Obviously that doesn't happen due to the workload. The GPU simply can't perform the work at the same speed when it has to process more pixels.
Similarly, it makes sense that for the SAME GPU the UI on rMBPs won't respond/perform just like the non-Retina models simply because of the additional workload that's a direct result of more pixels.
