Yep, GPUs are far far far from making 4K, or even 2560x1440/1600 playable with average setups. Yes, you can game at resolutions like that with the right settings with top-end GPUs, but with no AA you can still see jaggies easily, and with too much AA, you can crush even $1k+ of GPUs easily with a maxed out game.
After considering all of that, look at the average gaming rig (Steam stats, etc). People with even moderately decent GPUs are a distinct minority. The average 'gamers' PC is something in the range of a Core 2 Duo @ 3Ghz, 2GB of ram, and a 4850 or so.
Higher PPI will come in time, but it will be a fairly slow march. The major problem is that even if you show the average computer user a retina Macbook Pro, they aren't even very impressed, certainly not enough to spend a huge premium on it. Our market is absolutely dominated by the $300-$600 range of models for complete PCs/notebooks. Similarly in displays, probably 5,000 1080p and lower displays are sold for every 1440p/1600p screen, maybe even higher. The vast majority of stores don't even have 1440 or higher screens available at all. Even 1200p is about impossible to find.
We live in a lowest-common-denominator dominated world. Until that lowest common denominator is 4k (2025-2030?) I wouldn't expect to see a lot of movement forward. The best we can hope for in the interim is for GPU grunt for us enthusiasts to more effciently push higher resolutions, and for 1080p to begin to feature better AA and higher poly environments. Take the best looking game you've ever seen, and look at it in 2560x1600. Now go look at a good BluRay on a 1080p Plasma at a typical viewing distance. Do you see a bunch of jaggies? Which looks more realistic to you. Hence why 1080p has a ton more to offer as time goes by, even if I personally would like to see higher res AND higher PPI screens become more readily available at sane prices. Take something like BF3, give it 10x the poly count, add in better particle and color/brightness detail, and ~20x the texture resolution (so that even things at point blank range are still no worse than 1:1 pixel detail), and you'd have something that looked quite nearly real. So we're a few generations out from even being able to do that, that would require on the order of something at least 10 times as powerful as a GTX680 with 32GB or more of insanely fast video memory. Again, this is probably 10 years out, assuming we keep pushing the bar a bit at a time.