I dig broad skills, whether freeform (like Over the Edge or Risus), structured (like the forthcoming Leverage RPG) or somewhere in between (like yesterday's still-awkwardly-named baselines). Their benefits are obvious: you cover a lot of territory with very little in the way of rules or text. You get shorter character sheets that are usually easier to read since broad skills are usually written in plain language rather than expressed as arcane combinatiosn of skills, advantages and disadvantages. Plus, they tend to give themselves over to fast character creation. Compare the effort involved in creating a GURPS black op with just writing "Special Forces: 4d" and the benefit is obvious.
But that also speaks a little bit to the flaws of the approach. The most obvious is that you are sacrificing a certain amount of differentiation between characters if everyone on yous special forces squad has the same mechanical representation. This is less of a problem than it looks at first, though, since the creation of similarity usually serves to highlight differences more clearly. To use one of my favorite examples, all of the four musketeers have the common trait "Musketeer" but their differences (expressed in other traits like "Former Churchman," "Disgraced Nobleman" or "Ambitious") are all the more clear as a result.
More subtly, you can have problems with what PDQ calls the "penumbra" of these traits - the scope of actions and activities that fall under their auspices. To use the "Special Forces" example above, that probably covers a number of things, including fighting and sneaking, as well as knowledge of military procedure and protocol. But how does it differ from "Ninja"? There are color differences, certainly, but there is so much penumbra overlap that there's a danger that the color will get overwhelmed. The trick for dealing with that is, of course, to make sure the color stays important - so long as the traits are also suggesting game elements (special forces draws the military into the game, ninja draws in ninja clans, and in both cases the relationships complicate things) then it works out.
The last and perhaps most persistent problem is edge cases. Specific skills that MIGHT fall under the auspices of a trait, but also reasonably might not. For example, can our special forces guy fly a helicopter? It's conceivable, sure, but that's a pretty specialized skill, so it's also perfectly reasonable that he cannot. Making decisions about things like that on the fly can end up hurting a game if the GM has no external yardstick to use to make these decisions.
To solve this, we can breath some life into a very tired old chestnut in game design, specialization. Historically, specialization rules have existed for one reason: to make combat characters more scary, by allowing them to specialize in their weapon of choice, and thus deal disproportionate death upon all who oppose them. This is sufficiently lame that it's tainted the whole idea, but games with broad traits can benefit from the, but not in the usual way.
See, the first thing to do with specializations is to make sure the bonus they provide is small. They might make you a little better, but that bonus should be small enough that any min-maxer worth his salt would turn his nose up at it. And that's good, because the real purpose of the specializations is not to add a bonus, but rather to explicitly say "When the issue comes up, this skill falls under the auspices of this trait".
This becomes even more critical in a game with a fixed set of traits, like Leverage because that introduces a second issue. While some skills (like helicopter piloting) might fall outside any of the roles (Hitter, Hacker, Thief, Grifter & Mastermind) unless explicitly called in, you also have certain skills which could fall under multiple auspices. Take demolitions for example: it could equally reasonably fall under the purview of the hitter (it's military), the hacker (it's gadgetry) or the thief (blowing up safes, duh). The GM may make decisions about which role is appropriate on a case by case basis, but if a player wants to be good at demolitions under all circumstances, then a specialization under his role of choice effectively locks it down. That is to say, if thief is your best trait/role/skill whatever, you might specialize in demolitions under thief, not because it gives you a big bonus (it doesn't) but because it guarantees that when demolitions come up, you'll get to use Thief rather than some other value.
This also ends up giving a spin on some of the other problems, since you can now differentiate between the same trait with specialization without it being terribly abusive. We might all be special Forces (4d), but if the specializations around the table are medic, ECM, artillery and so on, it starts getting a little more G.I. Joe, and that's a good thing. That also speaks back to the cultural baselines from yesterday. This is still a half-idea, but if traits represent a broad culture, specializations can represent subcultures. That is to say, if you used earth as a baseline, a specific nationality might be a specialization, and in doing so would make the crazy Star Trek approach that earth equates to America a little bit less necessary. Yes, that introduces the idea that specialization might also be a limiter, (some cultures may not suggest the ability to drive a car, for example) but I doubt there's any strict need to adhere to that since, almost my definition, we're talking about interesting people. Such limits need be applied sparingly, if at all.
To bring it back around, I'm still very fond of broad traits, but using them definitely requires a firm grasp of their failings as well as their strengths. With the right tools, those failings can be smoothed over, and that in turn allows for a lot of power in very little space, which absolutely supports my particular flavor of GM laziness.
1 - More problematic is the difference between "Soldier" and "Special Forces" because the color difference is much smaller. The trick here is that the descriptors are implying a skill progression, much as if the traits had been "Warrior" and "Awesome Warrior". Since the actual awesomeness is usually in the trait values, this is one of those points that merits a conversation with the player to see if the distinction between their trait and another one is clear to _them_, and if so, have them explain it to you.
2- *Whistles innocently*