World texel density is limited by the actual size of the source dataset. They are not going to remake the entire megatexture in a higher resolution just for the PC users' sake, so don't expect more detail when you move closer to something.
The runtime memory load has been balanced so that all the texture information in any scene at 720p resolution will be the same for all platforms. The required texel information at this fixed resolution is limited, theoretically you'd need less then a single texel per pixel (although you need to multiply it because of normal, specular and probably some custom extra maps). As the camera moves around, some of the textures disappear from view so data can be discarded, while other tiles get into the view or get closer and thus have to be streamed in. Again, this is the point of using virtual texturing - they're not loading an entire 2K map for a single object if only 1/4th of it is going to be visible, or if it's far away from the camera.
This is why they can keep memory consumption at a constant and relatively low level. I repeat: they probably aren't even using 100MB of texture memory at all. That is the other and even more significant advantage of this technology, instead of the lack of texture repetition.
What the PC can do is to run the game at a significantly higher resolution, which would cause tiles further away from the camera to require a higher MIP level and thus increase the texture memory requirements. But tiles close to the camera are already at the highest MIP level and cannot load any better versions.
I really suggest reading up on the technology itself because there seems to be some misconceptions about it.