I ran into a similar quandry as you did. I'll tell you of an idea I came up with. It's a fun idea and I'm pretty certain it is unique, but I'm not sure it is feasible, as I have yet to build a fully-defined system for it.
My line of logic is this: in all of the gamist combat resolution systems I've ever seen, you could say that the combatants "follow" the clock. In other words, the clock "ticks" one unit of time, and each character involved in the combat must perform actions in order to catch up to that unit of time. This arbitrarily quantifies time, making it feel quite artificial, which has always bothered me. It occurred to me, why not let the players actually lay out their actions first and let the clock do the catching up, synchronizing the characters' actions as it goes.
The idea I came up with would be to sort the order of the characters in terms of who [b]declares[/b] their action first, [i]not[/i] who [b]acts[/b] first. At the very beginning of a combat, whether someone is even allowed to declare an action depends on whether they are:
(a) aware of the potential opponents, and
(b) interested in acting first, as opposed to simply reacting to the actions of the opponents.
Since (a) and (b) are mutually exclusive, we have three groups:
(1) unaware of combat,
(2) aware, but wanting to wait and react to the opponents/situation, and
(3) aware and wanting to act right away.
You'd only have to sub-sort these three groups. First, for group 1, you might roll or use the characters' awareness to figure out how long it takes them to clue in. This awareness time might be short-circuited as combat progressed. These characters would then filter into one of the other two groups. For group 2, the characters would simply wait. They could choose to jump into combat at any time (ie actively move themselves to group 1). The characters in Group 3 would act effectively simultaneously, but the players would declare actions in order of each character's speed (or agility, or some die-roll+dex, whatever) and have the character start performing those actions if not interfered with, such as by running into an invisible wall. Resolution would happen on the scale of the actions themselves, not on "rounds" or "turns".
Group 3 characters could not interfere with the "brief" actions of each other (except accidentally). They could interfere with longer actions, but only after they "notice" the action they want to interfere with. Until that point, they are committed to their originally stated intentions. For example, a holy knight could declare that he wants to go kill a goblin grunt. Next, the GM could declare that a nearby goblin-mage is preparing to cast a line-of-sight disintegrate spell. The holy knight's player would obviously be well advised to change his action to something defensive, such as ducking behind a tree. However, the knight is committed to attacking the goblin grunt for however long it takes the character to notice the goblin-mage, such as with a "Sense Motive" or "Awareness" check.
Group 2 characters suffer the natural time penalty - the opponent gets a head-start since the action is already in progress when the Group 2 character decides to react. Additionally, the Group 2 player must still roll a similar "Sense Motive" check as discussed above, if the intention of the action is not obvious. (Although if they fail the roll here, they are not committed to continue waiting, depending on the context; a wizard might wait to watch an enemy mage to see if he'll cast a spell, but decide to cast a "Detect Illusory Mage" spell when the mage doesn't seem to do anything significant after some time.)
In the rare case that everyone falls into group 2 right from the start, ie everyone waits to see what everyone else is doing before acting, you could do a roll to see who loses nerve first, and how long it takes.
Patrice, please let me know if these ideas interest you. If so, I'd like to talk some more.