This will only be a quick glance at combat, as without any mechanics... you get the idea.
Basically, I'm looking for a fairly adventurous, risk-friendly and fun approach to combat. It should be part of a game, rather than being the object of the game.
It strikes me that one of the major concerns, then, is that combat needs to be both simple and quick. I've experienced D&D 4E's evening-long combats between half-a-dozen on each side; that's really not what we're after.
To be quick, I want to limit both the die rolls and the choices involved. While there's also differences in power, one of the main drag factors on 4E combat is the tactical choices involved: people agonise over placement, dither over power selection, and sometimes discuss ability synergies for potentially several minutes on every single turn. In contrast, it's shockingly noticeable that in Arthur's AD&D game, a turn typically lasts ten seconds - even though we're still using a gridded map. Most importantly, this doesn't make it any less fun, and generally maintains the flow of play much better.
I don't mean that I want to constrain choice - what I want to avoid is choice paralysis, where you're scanning five or ten official options to work out the optimal once, rather than choosing "I attack" or coming up with something entirely original.
Another possibility for speed and fun is to crib the 4E idea of minions - very weak enemies who go down quickly. This perfectly fits the mood I'm going for, and speeds up combat enormously. It also creates a nice contrast between them and the keystone enemies that are central to the story, or have particularly fun and evocative abilities. Mathematically there isn't necessarily much difference between five battlebots with 10hp and fifty with 1hp and even worse marksmanship - but it's probably more fun blasting your way through the latter than chipping bits off the former, most of the time. That being said, I will be scouting for existing analysis of 4E combat before going too far with this idea, and it may not work out as well as I hope.
Second, risk should be small (as discussed previously). The injury system handles how defeated you are, not crippling injuries. You'll be able to escape later if the enemies capture you. Interesting plans should not be 'realistically' risky or fallible, because then people won't use them.
I'm thinking that whatever injury system I use should allow fairly rapid recovery, so that penalties from combat don't end up weighing on characters for weeks - especially as I don't plan to have healing magic or even insta-heal tech.