There sounds like there may be an excluded middle, here: You design the physics engine to deliver drama. Games do that almost automatically by the success/failure switch: there's ALWAYS anxiety about whether or not you will win, and catharsis with eradicating that tension.
In a "game mechanics as physics engine" game, the players have an incentive to take the maximum degree of preparatory steps to minimise that tension. You can see Gygax discussing it in the conclusion to his PHB, and Lewis Pulsipher talked about it a lot as well in old numbers of White Dwarf.
This leads to a game in which (for my taste, at least) preparation is over-rewarded, and hence becomes an excessive focus of play, and the goal of play-via-preparation becomes, in effect, to suck all the drama out of resolution. Scry-buff-teleport-SoD is the (notorious) paradigm of what I don't like about this sort of play.
Candy Land does that. Hell, craps does that.
I don't know Candy Land as anything but a name, but I don't think it is very intricate, nor does it have any preparation mechanics. Craps has no choices in play, nor any preparation mechanic (short of loading the dice, which of course will tend to drain away the tension around resolution - though perhaps there will be tension around discovery!).
An RPG is different in both respects - especially one with bounded accuracy. It is mechanically intricate, and it makes room for preparation.
I personally think that "fail forward", plus mechanical incentives for the players to not always want to use the biggest numbers possible (which can be achieved in various ways, including metagame rationing of the availability of the biggest numbers), are more reliable ways to ensure that the tension is retained. I prefer it if preparation is a source of colour (not mere colour, but a genuine way of colouring the consequences of resolution) rather than simply a tension-diffuser.
There's a lot of ways to solve the metagame problem of "combat is not dramatic/threatening enough" while still employing in-game logic.
For instance, instead of an arbitrary +2 bonus, you make it easy for monsters to gain Advantage on their attacks via monster traits or basic rules additions (flanking, fienting, high ground, better stealth rules, etc).
As soon as you are giving monsters the sorts of traits you talk about, or add in the sorts of rules you are talking about, in order to solve the metagame problem of "combat is not dramatic/threatening enough", you are admitting the existence of the metagame imperative. And then you want to give GMs advice about including choke points, flaking opportunities, high ground etc in their encounter designs - which, once again, is talking metagame. I simply can't see the rationale for pretending that the metagame is not there.
If you're saying that the mechanics have to also include ingame colour to explain them, well of course! That's not in dispute or even contentious in any RPG. The design question isn't whether or not there is colour, but (i) at what point is it introduced (by the designer, by the GM in the course of resolution, somewhere else?) and (ii) what are the constraints on its introduction (must it follow via ingame causal logic from earlier colour? or, for example, is a GM who is worrying that a rooftop duel is becoming boring allowed to simply stipulate "It starts raining heavily - your vision is obscured, and the tiles are very slippery!"?).
None of that necessarily violates in-game logic. Heck, even a +2 bonus doesn't necessarily violate in-game logic.
I don't quite see why you are framing the issue in terms of "violating ingame logic"? Very little
violates ingame logic. Of all the examples that have been mentioned, the only one I can think of is a 3E dragon's natural armour bonus (at least once the upper age categories are reached).
What I'm interested in is GMing advice that tells me how to set the numbers to achieve the desired metagame effect. That doesn't require breaking ingame logic - I can easily narrate around those numbers, whether via impromptu rainstorms, as per my example above, or via a list that tells me the various difficulties that correlate to various "objective" DCs (HeroWars, in its first incarnation, used this latter approach, as does Burning Wheel; HeroQuest revised uses the former approach, as does 4e). Bounded accuracy only makes this easier, by reducing the range of numbers that have to be correlated to "objective" difficulties.
What I don't want is a system that builds gnolls (for example)n on the assumption that they will use pack tactics and high ground to make their to-hit numbers work, and then doesn't tell me that anywhere, with the consequence that when I use gnolls they suck in play. Similarly, if leader NPCs are crucial for humanoids producing an interesting play experience, I want the game text to tell me that.
More generally, I want the game text to talk to me as a GM, trying to do stuff in the real world (namely, run a fun game) and not just present all its material from the point of view of the gameworld.
All a "Page 42" has to do is give you rough targets for what is achievable in-game at those given DC's.
<snip>
Assuming you gird your in-world logic so that it produces a satisfying metagame result, you can then come at it just from the metagame as a DM, if you want, without necessarily violating that in-world logic.
Your assumption is where all the action is. Designing mechanics that resolve ingame events purely in terms of ingame causal logic in such a way as to produce satisfying metagame results is non-trivial. Runequest doesn't do it. Rolemaster doesn't do it. D&D has never really tried to do it (unless you count hit points as ingame!), but where it has I think it's success has been doubtful.
The game designers think there is some rationale to assign levels to monsters, and to assigne them various statistics and XP values at those levels. Tell me the rationale! And then tell me how to generalise it across a range of other challenges and resolution situations! This is what 4e does. Bounded accuracy should make it easier, not harder, to do the same.
The problem arises when there's only metagame logic, and the in-world logic is an afterthought.
I don't really follow this. Boardgames have only metagame logic. RPGs always have fictional position within a shared imaginary situation - that's what makes an RPG an RPG. There are interesting questions about who is in charge of stipulating the fiction when, but when we're talking about page 42 we're almost always talking about improvisation, and hence on-the-spot stipulation of the fiction.
To give another example. A player says, "Is there any higher ground? My PC wants to move onto it, to get a benefit for attacking with his bow!". As GM, should I agree that there is high ground there, or not? And if there is, what sort of benefit should it give to bowfire (and, perhaps, how hard should it be to get to it)? The more the system helps me answer those questions quickly, and to smoothly adjudicate their implications for resolution, the better.
If to decide whether or not there is high ground, the system first requires me to have regard to the prevailing rock formations and water flow patterns in the area - information that I almost certainly won't have ready to hand, and don't know how to generate - then it will have failed me!
The BW answer would be along the following lines: Let the player make a Perception check - on a success, his/her PC spots the high ground, and a good success will let the PC get a bonus to making it there - perhaps by spotting a path, or seeing how s/he can head off the NPCs". BW doesn't set a default DC for the Perception check, but its version of bounded accuracy makes setting the DC pretty easy. Then there is a positioning check to get to the high ground, which is a generic feature of each combat round in any event. And then there is the +1 bonus for being on high ground - which I determine on the basis that +1 is the default bonus for any sort of advantage.
4e doesn't normally grant higher ground advantages. And not being a "theatre of the mind" system, it makes impromptue incorporation of terrain into the battlefied trickier. But letting a player make a Hard Acro or Athletics check as part of their PCs' move (using the skill appropriate to the terrain) in order to get a one-off +2 to hit foes surprised by their repositioning would seem pretty reasonable in the overall page 42 context. (I think [MENTION=64825]wrecan[/MENTION]'s article on the WotC website suggested a move action at paragon to gain combat advantage.)
This is the sort of thing I have in mind when I say I want advice that talks to the metagame, not just to ingame logic.