Just based on what I've researched, the optimal mining rate is achieved when you overclock your memory as high as it will go, and then adjust the core to a fraction of that speed. So if your memory reaches 1600, you might set your core to 0.7 of the memory speed (or some other factor, depending on specific card etc.) so set the core to 0.7 * 1600 = 1120.
Granted, 1120 core speed would be a decent overclock for a 7950, but the GPU is still capable of going further so you are limited by the 1600 memory speed, which itself is a very nice memory overclock. So to a miner, the video card is memory-limited because the GPU could be pushed further, if only the memory could go higher. This is all in view of the desire to maintain the video memory/core ratio.
I think a miner would want a video card that could reach a memory speed that corresponded to the maximum GPU overclock available - something like a "balanced" card where you overclock the core to something like 1250, and in order to 'balance' the card using a 0.7 ratio, you would need a memory overclock to something like 1790.
It's counter-intuitive, but apparently your mining rate will increase if you take an unbalanced card and lower the core speed to achieve the balanced ratio? But I'm still researching this, I haven't put it to use yet but will be setting up my rigs tonight or this weekend to start mining.
Could this new style of miner-focused video card be achieved without expensive R&D? Could you simply use a higher-speed memory chip that was better at reaching higher memory overclocks? If you are just substituting out video memory chips, I think it's a slam-dunk in view of current extreme demand for video cards, and I think some vendors already customize their memory chips between hynix, elpida 50, elpida 60, etc.