Unlike some others here I tend to stick pretty closely to the framework of N successess before 3 failures - though in time-sensitive situations I will incorporate a default 1 failure per round if the PCs don't make some sort of progress against the challenge (eg if the PCs are trying to stop an evil ritual during combat, every round that the ritual caster spends an action on the ritual, one failure is accumulated). My reason for doing this is that I like the discipline it puts on pacing - after N+2 skill checks have been made, we know that the situation has been resolved one way or another.
I think that skill challenges can be divided into two broad categories - complex skill checks, and scene resolution. A complex skill check is something like stopping a trap, or opening a door, etc. In this sort of challenge, you would normally be using a low-complexity challenge, and it will often be the case that only a couple of PCs are participating. Each check by one of these PCs will be making a clear contribution towards resolving the problem. You can narrate successful skill checks as straight forward progress towards the goal, and failed skill checks (short of the third failure) as temporary setbacks.
With scene resolution, on the other hand, the skill challenge is used to handle the resolution of a whole conflict. At least in my case, I use these sorts of skill challenges most often for social conflict. The key in this sort of challenge is to narrate the unfolding scene in a way that reflects the actions being undertaken by the PCs. (The skill challenge guidelines often present this in a very formalised way - eg with a successful Insight check the PC notes that the Duke is interested in being flattered about his family's past, which opens up the possibility of a History check - but in play I find this can normally be handled very fluidly.) The choices that the players make earlier in the challenge therefore affect the range of options they have later in the challenge (just as choices made earlier in a combat affect the range of choices available later in a combat).
As a GM, when you narrate the unfolding of a scene resolution skill challenge, you have to be definite enough that the whole thing is moving forward, but leave enough room for flexibility that you can incorporate the consequences of subsequent checks, and leave room to narrate the success or the failure that ends up occurring. (It's a bit like in combat - you don't want to narrate a PC who is dropped to 0 hp as having had his/her head cut off, given that the same PC might roll a 20 and be back on his/her feet next round - but you have to narrate
something or else it may as well be a mere dice game. Skill challenge narration requires the same sort of narrative agility on the part of the GM.)
Here are some links to discussions of skill challenges I've run in my game -
taming a bear and dealing with a water weird,
attending dinner at the invitation of the Baron only to find the party's mortal foe also at the dining table, and
persuading a captured prisoner to talk. Taming the bear is an example of a complex skill check, although the whole party participated, but the others are all examples of using the skill challenge to resolve a scene, with the earlier choices made in the challenge setting the parameters for the later choices, until the scene comes to its end (at somewhere between 3 and N+2 checks).