For three.js projects, you're able to view and design your scene from a browser. The default script shares the same camera to power both the browser view and the device view, but you could have two cameras if you wanted to.
Conceptually the important distinction between any camera in the browser in comparison to the camera on the device is that the device camera will get its orientation from what the user is looking at.
That's what's happening in this line of code:
if (vrMode) camera.quaternion.copy( frame.orientation );
To create a stereo effect when viewing a scene on the device, you can also shift the position of the camera.
All that said, the notion of cameras is inherently tricky in VR since in some sense the user becomes the camera. For now we've simply mirrored the way three.js works, but this is an area where we are really looking for feedback from developers and designers.