The site http://pokerslug.com (down now) had an interesting way to play - EV poker. Here's how it works: if both players are all in, instead of running out the rest of the cards, it just splits the pot based on equity.
So if you have QQ and the other guy has AK, both all in and the pot is 100, then it just awards you 55 and the other guy 45 (or whatever). Strategically it should not change correct play at all - since in the long term real play approaches EV play anyway. This way of playing just reduces the variance.
You need to be careful not to expose all the cards, or give exact EV values for match-ups on tables with more than 2 people, though, because you will inadvertently help figuring out the hole cards of players who folded. Plus, remembering that stuff, or trying to figure out people's folding ranges based on exact EV match-ups is painful. (There are some cases where it'd be pretty obvious - if 2 people all-in pre-flop with A6 vs KK, you would be able to tell if another player folded an ace based on the EV)
This version is a lot easier on the humans playing the game, though - it feels more fair, and you can play closer to what you feel is optimal, for more of the time. Unless you have no outs at all, it is hard to lose all your money.
Cash game perfect play in this game is the exact same as normal play - it simply has reduced variance. (There are other ways to modify poker that would increase variance, too - but people don't usually think they are better games - adding lots more random stuff, etc.) EV poker still has lots of variance. Bankroll-wise, I think there could be changes - you would be able to play with smaller bankrolls, and still have the same chance of going bust. Metagame (bankroll management) does change, though.
I'd love to see an ev-poker tournament. It would play pretty differently. (There would obviously have to be some kind of rule about how you lose if you money goes below 1sb (or maybe one ante). I think it would result in a much, much higher percent chance that the "best" players would win. The problem would be actually implementing it, though - it's very hard to do really. It could be simulated by just running every all-in hand 10x, though. To make it easier in person, maybe just run every all-in 5x. It wouldn't take that long. If it was online, it would be even simpler.