You are comparing old technology to new technology once again.
Eye Toy =/= Natal
DS voice recognition =/= Natal voice recognition
Voice recognition has come a long way, believe it or not. Did you play Tom Clancy's End War? Probably not. The entire game is played using voice commands, and as long as you don't stray from the words that the game instructs you to use, it works amazingly well. But that's using a crappy 360 headset microphone. Those things are cheap. Not as cheap as the DS's microphone, but cheap. Seeing Natal at E3 you can definitely tell-- it ain't gonna run cheap. The voice recognition was primarily demoed using Milo, because uh, most of the game appears to be you talking to Milo and forming a relationship. But yeah, from what was shown and from reading hands-on impressions, it works very well. Multiple press members got to speak to Milo and they will tell you-- he is very real. Peter Molyneux would first instruct the person to introduce them self to Milo. I didn't find a hands-on where the person wasn't greeted with their name in return. Everyone was greeted with a "Oh hello [enter name here]!" That's a lot of voice recording and a lot of voice recognition at work. Like you touched on, Milo senses your emotions through your speech. What other game does that? If you're sensing emotion through speech... that's gotta show something for how far voice recognition has come.
For the most part, I doubt the microphone is an issue with voice technology anymore - it's probably all in the programming. And no, I haven't played end war.
Now getting Milo to parrot your name back to you is nothing. As you said, all it has to say "Oh hello [insert name here]!" Getting Milo to say it in his voice is harder, but I suppose doable. But getting Milo to recognize the pictures you draw, unless they are preset, is simply impossible.
My entire point about Milo recognizing your emotions through speech is that no other game does that. If nothing else can do it, why should we believe that Milo, after a scant 3 months, has uncovered some miracle of AI where it can not only recognize everything you say with 100% accuracy, but recognize the emotions you expresss while saying it?
Milo is too intelligent. What we saw Milo do should not be possible, and this belief is furthered by the fact we saw so little of Milo. I do believe the secret to Milo is that it doesn't actually talk to you though - while it interacts with you, you aren't going to be able to hold a conversation with it.
Eye Toy is the PS2 equivalent to Natal IMO. So I am using Eye Toy as a point of reference to predict the success of Natal. Because I think for it's time, eye toy was probably just as revolutionary as Natal.
EDIT: I will add that just because something can be done does not mean it will be done. The wii-mote can add to the gameplay experience through fun and intuitive motion controls, but for the most part this simply hasn't happened. Likewise, Natal can be a revolution in gaming using it's unique control style to it's fullest extent, but I also don't see developers taking the time and effort to make it that, simply because they didn't put the time and effort into the wii necesary to make it's control scheme fundamentally better than the PS3 and Xbox 360 control scheme. Why should they changed their strategy and approach for Natal?