I think that battery will drain out more.
But, being capable of sending a real voice response to someone, in just one button push and forget style.... would be great!!
Beacuse now I have to check if the voice recognition did its work, or instead has written another message, one that sounds similar, but means different).
Also, I totally agree with f3flight's use cases, I think that, the posibilities for making the watch even more practical for its users, it's also great for developers, it just expands the scope of the apps that could be made. Audio is a great resource, and can have many creative uses and interactions. If possible "spy apps", are the only reason for not letting developers access to raw audio, it would be the only reason against thounsands of, really good ones in favor of having the use of raw audio available.
And like Marco has written, I also thought that, raw audio voice messages where going to be a given, maybe not with the first FW update, but I was really hoping for it!