In Solace, I'm actually going the opposite approach - instead of tackling the "dismounted soldier" problem by trying to make you more directly "there," I'm making it so that you're further detached, which tends to draw me in better. I find that MUCKs, text adventures, and reading books are much more realistic than "immersive" things. Then again, I also prefer a high-level approach to things as well.
Here's the interface I'm going to use if I ever get around to implementing more than just the renderer. :) Instead of directly moving your character, you give commands for motion and interaction and the like (just like in a MUD/MUCK/text adventure). In RL, you don't think about every single footstep you're taking, you simply approach a goal, like, "Okay, walk there, then walk there, then walk there, then pick up the orange." You only think about the low-level motions when you're learning how to make them, and to me, simulating footsteps and head movement and the like is just an expression of the low-level movements.
This command-oriented mechanism is also how I'm planning on defeating lag and the like - rather than require hideous amounts of bandwidth by expressing the low level, the (high-level) actions which are to be performed are broadcast within the room instead, and it's up to the clients to interpret the actions (through a client-side IK engine, or through the user's imagination if they prefer to just use a text-adventure client, or whatever). No, Solace isn't for FPSes. :) Also, all state is stored server-side, and can only be modified through relatively-abstract primitives (yay, implicit security model).
In any case. I feel that for more immersion (short of a braintap), you need less direct involvement, drawing the mind in, rather than trying to simulate the senses directly. The imagination is very powerful if you only let it thrive.
"Is not a quine" is not a quine.
I have a master's degree in science!
[ Hug Your Trikuare ]