A solution could be to use Augmented Reality when someone wants to type.
A camera (or two) set on the headset could record the real world outside the headset, then when someone wants to interact with the real world (type, grab snacks, or just move around) it could overlay, or replace the virtual world with the real world.
Another solution would be to project a virtual keyboard into the virtual world and then use a Leap Motion type device to detect where the user's fingers and hands are, and then match them to the virtual keyboard in world.They would type onto a virtual keyboard from the real world.
Lastly, you could use voice recognition to allow a user to speak the worlds they want.
Each system has its limitation and it pros and cons. However, the least likely to end up being used will be the ones that put the most strain on the user.