Performance per watt is generally only a major concern with Mid tier cards, in the high end generally speaking its a race to get the most powerful card out the door with all other factors being secondary, most hardware enthusiasts will look at speed first, then things like power usage, heat and noise second.
As the manufacturing process drops, the transistor size decreases the power efficiency goes up anyway so we get improvements in every generation, optimisations are always nice but I think I'd prefer the engineering teams focus on raw speed in the high end, that's the reason I buy the high end parts.