This would work much better. The main reason is latency. The added image will always lag behind the real world by some milliseconds (16ms minimum refresh latency for 60hz). There is also the image capture and processing latency which is considerable. This is true for both the Meta (hard augmented reality - hard AR) and the occulus rift (virtual reality - VR).

The difference in experience is for VR, the whole world lags so an added object will always be at the right place. With hard AR, the added object will be trying to catch up as the image moves.

From what I gather on google glass, they mostly aim for soft AR. This is where you add information on top of the real world but it doesn't map directly to a location in the real world: if you move your head, the added image stays the same. This is much more resistant to delays and could be useable for extended periods of time without discomfort.

For a better and more technical writeup:

http://blogs.valvesoftware.com/abrash/why-you-wont-see-hard-...