Open world games/games with distant terrain that you can explore/approach use LOD of varying quality, so, while it's very demanding, it's not "that" much more demanding than a linear game because the assets/models/content in the distance aren't rendered in high quality until you approach it (Ideally.). LOD quality is whatever quality the Xbox360/PS3 can handle (There tend to be several different levels of quality depending on the distance. I think I heard somewhere that 7-8 different levels of LOD quality for trees, for example, is normal for an AAA game.) It's a continuous process as you move about the map.
Uncharted 3 has the explorable content (All levels combined.) of, like, 5% of Skyrim (Example.), and they probably spent a long time developing that. There is a lot of detail in those areas.
Unless you have a team of 10.000+ talented people working for a couple of years, I don't think it's possible to create such a world. And teams of that size are very inefficient (Atleast in a complex production like this, where everybodys result needs to be put correctly into a single product; the game.)
I think procedural generation is the only way to make something resembling a Skyrim size world with Uncharted 3 detail. But to even get to that point, unless someone brilliant comes up with a perfect algorithm/solution, the gaming industry needs to start experimenting more with procedeural generation and pool together solutions, and iterate their way to a solution/method that can make something like a Skyrim+U3 detailed world.
You know, like a program that automatically makes a lot of the models needed, places them around the map in a logical manner, and automatically generates the necessary textures and puts them, logically, on the models.
There are already programs like CityEngine out there, which generates entire cities on the fly.
Procedural generation and manual work on the content generated is probably the only way to make Skyrim like games much more detailed/larger/varied in the future.
Bethesda already had most of the engine developed, and they still spent several years on the project. Games will probably become much more expensive to make using the same old method of development, so I think it'd be much more efficient to develop worlds using procedural tech in more areas of development (Open world games usually use some form of procedural generation to generate trees/generate initial heightmaps perhaps, and to a much lesser degree, texture placement over large areas I think.)
And/Or I really think they should consider discussing something like a universal game engine at one of those developer conferences. It's very unlikely to make something like it at the moment, but I think they should atleast talk about it.
If there was an open source engine with everything necessary to make a game integrated (3D modeller, 2D software like photoshop, texture/heightmap editor/generator, open for all scripts, physics, A.I, etc.) into it, making games would probably be much easier. Developers wouldn't even need to spend a lot of their budget on making an engine, or licensing an engine.
And, they could potentially pool together all the resources/assets in a shared database -- if a huge amount of assets/models/content/worlds were readily accessible, a prototype game could be built really fast. Then they could modify the content so it doesn't resemble anything that's in the other game (From which they borrowed the content/models/assets.), or make new content.
Not sure how viable that is, though.
I'm not saying it should be exactly like that, but a shared engine with everything necessary to make a game is the general idea. It'd probably cost like 1$ billion dollars to make at first though.