Well, it depends on what you're trying to achieve. But there is a difference between a description of an NPC as "cowardly", with the GM adjudicating behaviour on the basis of that, and the assignment of a morale number to that NPC a la classic D&D, with the GM making morale checks at certain specified intervals.
Why would you go the second way? Maybe you want to make it easier for the GM (as LostSoul describes). Maybe you want a mechanic that the players can then directly affect (eg by having their PCs cast spells or use abilities that grant bonuses or penalties to morale checks). GM fiat as a resolution sytem doesn't generally encourage players to build up, and expend, mechanical resouces in relation to in the same sort of way.
For me, the "unexpected results" aspect of social mechanics is the most important. That's why I personally enjoy a skill-challenge style resolution system, where the mechanics oblige the GM and the players to continue to introduce new content into the situation (via further skill checks from the players, and further narration of the resolution of those checks from the GM). That new content, as it feeds on earlier-introduced content, produces results that no one at the table anticpated or was aiming for going into the situation.
I don't think the GM is an adversary of the players. But I do think the GM, at least traditionally, has the role of providing adversity. And I think it becomes harder to do this if the GM is also in charge of determining, without mechanical guidance or constraint, the consequences of the players' actions and choices.
In a system without morale rules, for example, a GM who has the monsters break of combat and flee may be doing one of several things. Perhaps s/he is saving the players from the risk of a TPK. Perhaps s/he is deliberately adding to the adversity of the situation - now instead of cutting them down, you have to chase them and cut them down. But I don't think s/he can easily and plausibly say "I was just playing my NPC/monster accordig to its character - I'm not responsible for how it changed the situation."
Mechanics are a device for divesting (or, perhaps, pooling - because everyone has defined ways in which they can affect the mechanical resolution) this sort of responsibiilty for outcomes.
I'm not talking about "guard rails", and I think that your reading of me in that way suggests that we may think about the activity of RPGing quite differently.
I'm interested in responsibility and surprise. As I said above, I want unexpected things to happen. For unexpected things to happen, it has to be the case (I think) that no single participant in the game gets to decide, in a definitive fashion, the way that any given interesting event resolves. (Boring events, like restocking iron rations, I'm happy to pass over very quickly and just let the players make the appropriate changes on their character sheets.) Good mechanics, by this criterion, provide resolutions that aren't at the dictate of any single participant (including the GM).
The issue of responsibiity is related. If the GM is responsible for how things resolve, then the GM is in the awkward position of having both to pose the challenges - be they hordes of orcs attacking the town, or uncooperative bureaucrats stopping the PCs finding out what they want to know - and deciding whether or not the players attempts at having their PCs overcome those challenges succeed. I'm happy to take responsibility for posing challenges, and if the challenges fall flat that's on me. (Part of the measure of a good system, then, is the support it gives me in designing challenges that won't fall flat - this is related both to build rules, and to certain aspects of resolution rules.) But in general I don't want to take responsibility for whether the players succeed - I don't want that conflict of interest. The players' success or failure is on them and the dice.
That's not to say that I'm never going to exercise mercy. But generally, I want that mercy to operate at the scene-framing level, not the action resolution level. For instance, the one time in my 4e game there was a TPK, for example, I discussed with the players who wanted to start a new PC and who didn't, and those who wanted to keep their PCs began the next session with their PC regaining consciousness locked in a prison cell. (The 4e rules permit this, because they don't equate 0 hp with death.)
But it is clear to everyone involved that this is not an application of the action resolution mechanics. The mechanics having been applied, and the scene having been resolved (the PCs were soundly defeated), I have framed a new scene which builds logically on the previous one within the parameters permitted by its resolution. And there is no conflict of interest in framing a scene sympathetically rather than harshly. In fact, I think a big part of learning how to GM well is getting a sense for what scenes are best framed in response to the outcome of prior scenes. (This is part of avoiding the framing of scenes that fall flat.)
There are complexities in the above - for example, is deciding who a monster attacks in a combat best conceived of as an exercise in action resolution (to be settled via the movement and reach rules plus GM fiat), or as a little mini scene-framing within the broader scene? I do my best to treat it as the latter rather than the former - so to treat it as a reframing of the scene, with the goal as GM being to make sure it doesn't fall flat - and then, having framed a scene, to play my monster all out and do my best to wipe out those pesky PCs!
