The fluff is there to help you as the DM build a coherent fictional world.

First, without any explanation even to the GM, "there's a bearded devil with an intelligent glaive in a cell" seems to not make a lot of sense. "But why didn't they disarm him?" "These guys are devils, why is a devil in a cell?" etc. PCs tend to investigate things and want to know "why," so many times when they do I go to the fluff and let them uncover some of it, it builds the sense in them that the world is a logical one and it's not, as in 1e sometimes, "a huge dragon in a room with only 5' corridors going out of it." Frankly, even if they don't ask, by comforting you the GM that there is a reason (instead of just having a list of rooms with a monster listed in each) it makes you more confident in the game's fiction as well.

Second, they try to do this when there is a chance the PCs will interact with a NPC. In this case, they are in a cell, and can be spoken with without opening the cell door (in fact I am not sure why someone would open the door...). In cases like that, it's nice to have something to hang your hat on. (Not to harp on this specific encounter because you're asking in general, but it just says he's "mad" and "full of rage," it does NOT say he attacks the PCs on sight, which means he could potentially be enlisted as an ally against the villain - a crazy, evil ally, but hey). And don't forget they then get his intelligent glaive, which was a witness to those events. "Hey glaive what was the deal with that guy? "Uhhh...." Or maybe they capture someone else, a guard or whatnot, or they free someone else, and ask them "what was the deal with that devil down there..." There are many, many mouths information like this can issue from.

Third, even a detailed AP is meant to only be the starting off point for your own game. You are expected to expand on it, change it up, mash it up with other stuff, alter it to incorporate your own plots and characters, etc. By adding a little more info to the characters (in the case of the Howling Fiend, the fluff is exactly two sentences long) it provides a plethora of additional expansion hooks that a GM can use during adventure planning. When another devil shows up later, maybe he's not just a random encounter, maybe he's looking for his buddy Szasmir who went MIA a long time ago and now the PCs have a glaive that reeks of him.

Anyway, the actual killing of monsters and NPCs forms a reasonably small part of the overall action in a game of Pathfinder, the investigation and roleplay and all that form a huge part as well, and the fluff is there to fuel it.

Thou shalt not sacrifice caster levels.

As the golden rule of 3.5 multiclssing, it's hard to beat the above. It's quite possible to apply requirements gathering methods to 3.5 as well as 4e, and estimate approximate to-hit and damage at each level.

Plan out your character to 20.

To answer:

s there an effective way to identify class combinations which seem appealing up front, but which will fall behind as the characters level up (without actually taking the character through its paces in a game)?

is simple: plan out your character to level 20. Identify, at each level, what benefit the character is deriving from your choices. Test against the requirements given in your requirements step. This way, when your character shows up in game, you have an idea of your intent and the capabilities of the character.

To answer what classes synergize well:

Non primary-casting classes tend to synergize well.

Given that most aspects of a level up are cumulative with prior choices (BAB, feats, HP) you want to avoid class features which depend on your level in the class. Therefore, avoid class features which have level as a variable within the feature.

Rages from barbarian are fine. You get more as you level up:

a barbarian can rage for a number of rounds per day equal to 4 + her Constitution modifier. At each level after 1st, she can rage for 2 additional rounds.

but you don't have the benefits of the rage reduced (proportionally to your compatriots) as you see in spells.

For the casting classes it's certainly possible to combine them in "gishy" characters (spell-slinging fighters) but they require a specific focus into the combination, preferably enabled by appropriate choice of prestige classes.

So, if you fail to raise rogue, you don't get as much sneak attack bonus damage, but if you're increasing fighter as the other class, you get a higher BAB to compensate.

Whereas a rogue would effectively be writing off "improved uncanny dodge" due to multiclassing.

Therefore, a good combination can be fighter/rogue, because each class brings a separate thing to the table, without having a significant opportunity cost for choosing the other. Other good combinations are those that lead up to exceptional PrCs, which combine features from both classes in useful synergy.