One of the biggest things, I think, that games do wrong is that they leave a lot of space where there's nothing particularly interesting to do. The other mistake, of course, is to give the players too many things to do, and that's just as bad. I'm thinking about textbook combats in older-edition D&D games and combats in 4th edition D&D. It's easy to "zone out" in either of them, and kind of sit there, attacking the same target until it's dead.
|An Igniter, from Bloodline Champions.|
The other big thing is that there's always a mind game going on, where you've got to predict what they're going to do to make your own move optimal, and to attempt to out-play your enemy. All of the combatants are roughly equivalent in ability and in power, so it's up to the player's decisions to make the difference. You can't rely on brute force (at least, unless you're a class that enjoys a size/strength/durability advantage), and even if you try to, you need to consider your enemy's moves and their likely counterattacks.
It's something that should, and could, be codified into a neat game mechanic, and I think it's where my attempts at development will go. I would love to bring yomi (in the David Sirlin sense of being able to read your opponent) into roleplaying games. It's the one part of any game that I truly enjoy, and it should be present in one of my most long-lasting hobbies.