Manbearcat
Legend
See, I guess where you see marginalize, I see deemphasize. And again, I think this is a function of player-type, not system. Even in old-school games, most players I met would simply attack, or look through their magic items. Spellcasters would look over their spells. I've never played in games where players were always looking for chandeliers to swing on or fireplaces to push people into.
The best way to get players to improvise, I've found, is to give them upfront knowledge of the outcome and that attempting to engage with the environment (which is really what improvise means) is their tactically most sound option
I wanted to drop in right quick to address these two points.
It seems that mostly what we're discussing here is the propensity to stunt.
In my 1e games, who was "stunting" the most? The Thief and the Monk. It was quite easy to discern why they were stunting (trying to engage me to make favorable rulings and/or provide transparent odds for success on their wacky stunts) while the Fighters were continuously pressing the "meat-grinder attack button" while the spellcasters were looking through their spells and then pressing the appropriate "problem solving spell button." Its because their default options were terribly ineffectual and engaging in symmetric warfare meant death.
I'm going to fast forward to my present game.
I have a Thief (Rogue) and a "Monk" in my present game. The "Monk" is a Bladesinger who is basically a Jedi with a FIghter/Mage skin. He uses the Bladesong (Force). They are both light armored, mobile, duelists/fencers. One is an Errol Flynn Swashbuckler and the other is a Jedi Knight basically. These guys have a fair bit of functional and thematic overlap due to the nature of the archetypes and how the mechanics go about supporting that:
1) They have above average, passive AC for their level (with the Bladesinger at Defender levels).
2) They have activatable abilities that buff their defenses significantly (especially AC and Reflex). They both have enough of these that they can sustain these defenses for 80 % (or more) of the daily percentage of combat rounds.
3) They both do significant at-will, single target damage (The Bladesinger just under and the Rogue just exceeding of-level, high damage expression).
4) They both have a robust suite of off-turn actions to riposte, deny/subvert attacks against them, or move in response to a trigger.
However, while one gets by on the "aerialist swashbuckler" shtick, the other gets by on "using the force (Bladesong)". When one stunts (invoked p42) to engage what might be an otherwise benign terrain feature, he is going to be deploying his ridiculous Acrobatics (automatically passing the Medium DC and 80 % on the Hard) and doing the sliding down the bannister into a leaping kick, knocking the Ogre into the brazier. The other also has a ridiculous Acrobatics check, but he is much more likely to leverage his "use the force shtick", deploying his ridiculous Arcana (automatically passing the Medium DC and 80 % on the Hard) and force-throwing the brazier (with its burning coals) at the Ogre.
Do they stunt as much as 1e? No. That was literally an every round affair. Do they stunt a lot? Absolutely. And when they do, their stunting is informed by the (i) the mechanical constraints of their PC build tools, the (ii) thematic guidance of their archetype, and (iii) the tier-based, genre credibility test that [MENTION=42582]pemerton[/MENTION] was speaking of.
One archetype is (tactically, not strategically) stunting more than ever in my 4e games; spellcasters. They are deploying Arcana, Nature, and Religion with interesting combos riffed off of the baked-in thematics and mechanics of their respective power sources and spells far and away more than I've ever seen before.