One of the key differences are drivers/hardware-fixed function stuff. Its completely different generation, take those same desktop cards, or cards with similar power made up to date on current processes with upto date optimisations in architecture, etc, they would use insanely less power. 3-4 process node drops is a huge saving in power and die size, a 7900gtx would likely be below 50mm2 with smaller mem bus and efficiency savings by now, keep in mind a Tegra 3 is 80mm2, and provided **** performance in that test in comparison(obviously with cpu cores included though and the 5th stupid core). Arm chips aren't nearly as small as people think, with both A15 massively increasing size along with the performance, and power, while the 3mm2 type numbers people talk about are the individual cpu core sizes without the rest of the chip, AMD/Intel cores aren't that much bigger with bobcat/atom.
Mobile is power limited, it was growing fast because they weren't at the power limit more than anything else, how they are hitting the power wall. Though a15 can itself use a pretty hefty amount of power compared to mobile chips 5 years ago, its about using that power, doing something quicker then turning to idle with WAY lower idle usage than older mobile chips. But ultimately those chips only had a generation or two of max power usage growth before they hit a brick wall, and thats basically whats happening. They've grown the core size significantly, and overall chip size, but because a mobile battery can only take so much load power before the device eventually becomes useless(because people want more than 2-3 hour battery life), they are now bound by the same growth limits as AMD/Nvidia in desktop, every new process might mean you get 70-80% performance from the same die size in the same power limits, doesn't matter if that is 80mm2 and 3W or 350mm2 and 250W.
The next big jump in mobile performance will come from battery technology, by jump in performance I mean gaining from mobile towards desktop card performance. Till you can use 15W load in a tablet without the thing overheating or battery running out in 5 minutes, they're stuck. Thing is, this was patently obvious years ago, its EXACTLY what happens in desktop.
For years the discrete gpu's grew in size and power limit, that size/cost/power limit is now and has been for a while around the 300-500mm2(depending on market segment/cost) and power grew up to 300W then backed off a bit, die size is now backing off as well due to increasingly complex processes hurting yields and costs of very big cores.
In 5/10 years you'll be able to buy a mac with powerVR graphics Pottsey 
