Let me agree with you in a very long-winded fashion.
As I see it, a basic and necessary conceit of D&D is that the challenges played out at the table are almost always those that are within the party's grasp, plus or minus a bit. Otherwise, D&D fails as a game: neither regularly failing at the impossible nor regularly succeeding at the trivial is a good use of time. I think this is a necessary goal to pursue so that D&D is a fun game.
On the second hand nearly everyone finds verisimilitude (or world engagement or whatever you want to call it) important to some degree. I don't think this is a necessary goal to pursue for D&D to be a game, but it probably is for it to be a role-playing game.
There are two basic approaches one can take when pursuing these two goals. First, one can declare that all challenges have particular probabilities (+/- a bit) regardless of what the characters are actually doing in the world. This is party or game primacy. Second, one can say that what the inherent qualities of something which can be done defines everything about the challenge, and that the mechanical capabilities of the party are essentially irrelevant (+/- a bit). This is world or verisimilitude primacy. In practice, of course, a system will be somewhere in the middle.
Like you, I tend to think 4e went a little too far on the gamist side of things, for basically the same reasons given in the RPG.net post.
I think the solution is to keep part of the best of 3e and 4e. Each was nicely consistent in one area, and poor in the other. 4e was provided good recommendations regarding the math of challenges, but no consistency as to what those challenges actually are. Meanwhile, 3e provided copious examples of specific challenges, but the qualitative feel of the math could be hilariously inconsistent from check to check. In fact, that along with the ridiculous stacking issues pretty much made the 3e skill system almost impossible to extend in cool directions. (Any skill-based casting system, for example, definitely required a gentleman's agreement not to abuse.)
The nature of the d20 impacts the specific numbers, of course, because it defines the limits and variance of what is possible. My favored rule-of-thumb is that a maximum level, maximally buffed, and maximally focused character can essentially just barely surpass on a 1 what the average untrained commoner can do on a 20. In other words, the skill system would permit DCs from 0 to 40, and pretty much nothing else. Multiples of 5 would define tiers of difficulty in an absolute sense, and be a useful guide for putting down challenges when you don't specifically care about the "relative difficulty" to the party. This is, in fact, exactly the table listed in the 3.5 skills chapter, but it was adhered to very poorly.
I think keeping both absolute and relative guidelines is important both before and during play. Before play, one can worry about either as much as one wants, building the world first or the challenge, or anything in between. During play, the big difference comes from elements that have already been described, versus those that have not. Elements that have already been described can use the absolute guidelines (say at some point the PCs wonder if they can climb the already-established rain-slick cliff) while elements that have not been described (like the PCs ask if there is a cliff they could climb) could use the relative rules if the DM want that to be an avenue forward, and then determine the sort of cliff it is if that becomes relevant.
Finally, the DM is always free to ignore these examples of absolute difficulty. One can still run the game in the 4e manner, in other words. (And in a modular 5e, jettisoning them will hopefully be less problematic with rules lawyer types.) But for those who want greater emphasis on a verisimilitudinous world, I think it is a lot harder without the absolute guidelines than it is with them.
