I hate level scaling - I've seen it done well, but only in games with heavy strategic depth and where the games themselves either strongly justify it or otherwise are straightly mechanical.
In most cases, level scaling is an example of bad design, plain and simple. Sometimes the design is intentionally bad, like Oblivion - it lets them focus on other things while pretending to give the player a meaningful progression. It's still bad design, though. Especially since the effective result, for someone like myself, is to insure every area of every region of the game fails to provide a challenge - there is literally nowhere I can go and nothing I can do that isn't boring and monotonous grind the way many of these games handle it.
There are plenty of good ways to handle progressive difficulty without resorting to scaling, even in open world games. And there are better ways to handle scaling than the way most games do it.
1) Have strong enemies that the player can successfully escape from. This lets them learn which enemies they can handle and which they can't. In "strong" areas, you don't need to make every enemy strong - just have enough strong enemies that it becomes a risk/reward thing. Can I get in, get what I need, and get out before I draw the wrath of the Chaos Dragon that roams these caverns? This is even better when the player can put effort and preparation into figuring out ways to avoid conflict - a temporary camoflouge potion, having the enemy be sound focused, having the enemy generally ignore the player unless the player does something stupid and provokes it, have the player pack a bunch of temporary slowing items to give him a chance of escaping if conflict is triggered.
Later on, the player can return and vanquish the enemy in combat if he so wishes, perhaps for those rewards he couldn't get before. High level enemies don't mean the area needs to be completely closed of to him.
2) Low level enemies don't bother the player any more. Have them run away from a strong looking player. In fact, have different behaviours for different enemies that are variants of this, with the key being that players don't need to waste time on monsters that are to weak for them (unless they choose to hunt those monsters down, of course).
3) Have story justified "scaling" based on WHAT the player does, or things that have actually CHANGED in the world, not on how much experience they've gained. For example: The ever present hit squads. If the player killed the last squad with greater fireball, send warriors with rings of fire immunity as his reputation expands. As the player becomes more of a thorn in the side of the local power structure, they dedicate more resources to stopping him, focusing on those units and tactics most effective against him previously. Local bandits who've heard of the hero would obviously be more prepared - you don't even need to scale their levels to make them more of a challenge, so much as modify their build to counter the player. Maybe to such a point the player eventually needs to adjust his strategy rather than doing the same thing over and over again. Basically - rather than scale, /adapt/. Then, when the player inevitably games the system, it feels justified in game - he outsmarted his opponents by misdirecting them as to his true power, after all!
4) It should be visually obvious, in most cases, which enemies a player should consider dangerous and which he can handle. Let the player choose when he wants to tackle tougher enemies, and to avoid those big lumbering beasts that are clearly more powerful than him. Not to say this can't be switched up now and again...
There are plenty of other options games can take - most just don't bother, preferring to fake it and save time and money for something most people don't care about.
Every game I've played with level scaling would only be improved by removing 95% of encounters completely.