No one is using a 28nm process in a handset right now. Apple is waiting on 28nm fab to come fully online so they can make an LTE radio for the iPhone 5 that won't suck down as much battery juice. In one way they are smart to wait for it (everyone hates short battery life) but they are also risking further marketshare erosion (because people make purchasing decisions based on more obvious upgrades, and also LTE is fast enough that many will sacrifice battery life to use it).
Apple will probably be first off the line with 28nm SoCs, but they aren't there yet. The A5 apparently has not only the beefier GPU, but some other custom circuitry, like a fully pipelined FPU. If you guys want a good write up of it Anandtech generally does the best overviews of the nitty-gritty of SoC design, like in their
iPad 2 review.
Edit: It's also worth noting that it's not so much that the Tegra 2 GPU sucked, but that it was almost a year late in shipping. What would have been cutting-edge in Q2 2010 was sort of meh in Q1 2011.