I had a pretty large sample size, sitting through 5 demos, capturing about 40 different movements from a variety of journalists. Across those 40 movements, the fastest life-to-screen transition was .08 seconds, while the slowest was .12 seconds. A tenth of a second was the consistent average, though.

What that means is that, in its current state, Natal is not instant, one-to-one motion capture. It's close, very close, but once you start playing you'll undoubtedly notice a tiny delay. It's to be expected, certainly. Even high-end Hollywood studio motion capture devices have a slight delay, and a camera sitting in front of your TV doesn't come close to the fidelity of 30 cameras in a massive studio.

Interestingly, it would appear that the demo being shown in New York is identical to the one shown at E3, right down to the delay. I wasn't allowed to film the screen, but this exclusive CNET video from last August does feature person-and-screen footage and it gives a great sense of what you can expect, delay-wise. The appropriate footage kicks in at 2:05: