As I see it, the sorts of issues being discussed in this thread arise when the narrative element and the mechanical element come apart.
For instance, as I understand [MENTION=5834]Celtavian[/MENTION]'s complaint, the narrative element of the Marilith promises something that the stablocks don't deliver. [MENTION=58416]Johnny3D3D[/MENTION] had a similar sort of experience with 4e (fortunately for me, this hasn't happened to my group).
When this sort of dissonance occurs - for instance, the flavour text says that the Marilith is such-and-such a thing (say, a threat to all but the mightiest of heroes), and then in play it turns out differently (the ongoing ficiton of the game isnt framing the PCs as the mightiest of heroes, and yet they easily defeat a Marilith with really very little effort at all) - then what is to be done?
I don't think just telling those experiencing the dissonance that they're doing it wrong, or that they're metagaming too much, is very helpful. (Which is what some posters in this thread seem to be saying.) After all, if the mechanics and narrative complemented one another, then this sort of metagaming wouldn't matter - it would just reinforce the narrative.
My own experience, in the play of my own game, is that "mystery" in the sense of how many levels/HD, or damage per hit, is not that big a contriubutor to the drama of the game. The sort of suspense that I personally find more effective is suspense as to motivations, or (in combat) as to the broader tactical situation (eg are there reinforcements?).
An example of the sort of tactical uncertainty that I have found works for me is in this post - what starts as a fight against hobgoblin wyvern riders - with tactical uncertainy resultiing from the separation of the PCs on the battlefield - escalates as a hobgoblin phalanx marches down from the ridge, and then the hobgoblins unleash their chained chimera, and then the three-headed firedrake Calastryx flies onto the scene to find out what is being done to her child, and then - after defeating Calastryx - the attempt to harness her magical power attracts the attention of a pack of mooncalves.
A broad knowledge of a creature's capabilities can often enhance this sort of supsense - if the players know, for instance, that a creature has the ability to blast them with a gaze and reduce them to 0 hp (in 4e, bodaks have this ability against a target suffering the weakened condition), then they start having to plan around avoiding getting blasted in this way, which becomes a source of doubt and threat as the encounter unfolds. Other times, I find it better to leave things hinted at - eg when the PCs fought Calasytryx they knew that there would be some significance to her having three heads, but they weren't entirely sure what that would be - they got a shock when she grew another head rather than having one chopped off! This is a case where I wouldn't casually share the stat block with the players (although if a player used an ability to learn it, I would dutifully tell them).
For my group, the framing of a situation such that fighting will surely lead to the PCs' doom and I, as GM, make that clear to the players, probably wouldn't work, because it would be putting the GM rather than the players too much in the driving seat of the plot. If the PCs have overreached, I prefer that to be something that emerges out of the mechanical resolution of the situation. (Eg in my Burning Wheel game, a PC was in the catacombs under the city of Hardby, hunting down Chemosh cultists. At first he was beating up on them, but then they got the upper hand, and so he was taken down - and regained consciousness inside an iron maiden, which was then the starting point for the next session.)
I tend to find the mechanics, and the story/flavour they give rise to and that hangs around them, shape this. So the 4e games tend to be have a fair bit of combat, as 4e tends to emphasise combat as the ultimate mode for resolving conflicts. But my RM games in the past, and my BW game, have less combat, as the mechanical resources that the players have for non-combat, and the systems to adjudicate these, are just as robust as the combat mechanics.
I certainly have combat-free or sessions in 4e - but these tend to be the outcome of player/PC motivations in relation to the session, rather than of judgments about whom the PCs could or could not successfully hope to kill.
