The 360's setup is actually not all that similar. On 360 only a part of the gpu had access to the edram, and it was automatic, the developer didn't have to do anything to use it. It did require some custom work to reduce the hit of 720+AA, but in general, developers simply decided not to bother with that, resorting to lower resolutions or no AA if they surpassed the edram size... But even then they benefited from the memory setup.
On durango the entire gpu has acess to the esram, and some orchestration of data is required to make use of the setup... There are some hardware that can help orchestrating this, but only Ms knows how automatically it can actually help.
This very leak implies that the command buffers that the gpu reads come from the data move engines and not directly from the cpu (the DME patent also implied that one of the operations would be to copy data directly from one processor to another)... That alone may require some effort for the developer to synchronize the workflow, otherwise you could end up with both the gpu and cpu stalling waiting for the other.