anexanhume
Veteran
I think you could get a GPU with similar if not much better power numbers than a laptop part by doing a custom production. Right now I believe the laptop parts are down clocked and screen for best power numbers from the same production runs (as the full 7870 for example). All those parts are on the same process variation (likely 28nm HP as Nvidia is using) and targeting ~1 GHz performance.
If for instance, they chose to build a part with the same number of shaders and targeting only 750 MHz with a custom layout, they should be able to achieve higher density (thus smaller die size) as the transistors don't need to be as large. Larger transistors give you stronger drive and faster swiching times. Reducing their size will also lower leakage.
Going to a slower clock may allow them to use a LP variation of the 28nm process. That should further help power consumption and leakage.
With customization and a lower clock (750 MHz), I think you could reach a GPU a size of around 175mm and about 2 Tflop performance (1280 shaders & 750 Mhz) in about 75W. At least I hope. Another 35-45 for the CPU, 20 for fast RAM, and 20 for the rest of the system and you have a system around a 150W power requirement. Not unreasonable.
I'm going to guess migration of an existing design to an LP process won't be peas and carrots. Designs tend to be optimized for the process they are on.