Definitely not. I don't believe that voice recognition will rival the keyboard in the near future, probably never. I find it difficult to concentrate when talking to a PC and it feels rather unnatural and ridiculous. If you need a short pause for thinking, you just stop typing. Speech recognition, however, urges you to keep talking consistently and without "umms" and "mmhs".
I'm not saying accurate recognition of speech is/will be technologically impossible, but I think it's rather inconvenient and unnatural, at least for text input.

I think we'll always want a silent method of input because talking loudly is socially stressful (even subvocalizing). Some time in the future I think that it'll become more of a mindlink thing, but that's far off. I work with functional brain studies (fMRI) myself and have read about MEG some so I have a bit of an idea about this. Not in the next five years I believe. Maybe not in the next twenty for all I know, although there's a lot of research going on at least.

I also believe that we'll always be using our hands. They're such wonderful instruments. Those keyboards where you can run macros and effects by moving your fingers in certain patterns are interesting.

YOW!    that video is hilarious !!!!

  It would be nice to see someone attempting to demonstrate Dragon Naturally Speaking in the same way for comparison since they appear to be the gold standard.

Proof of concept for programming via voice. 

The greatest advantage (aside from the obvious ones, of course) is the large number of "quick sounds", like "oink", "boink", etc. This allows for many more primitive commands to be mapped, which would otherwise require many keypresses on the keyboard.

Error correction seems a much smaller deal than you'd think at first, partially because programmers tend to be reasonable about variable names, etc.

General disadvantages for voice coding:

• Requires learning of an essentially new language based on monosyllabic words.  Though, as voice recognition becomes more common, I predict that these will become standard in the next few decades.

• Still inefficient for arbitrary words - better hope yours' are in the dictionary.  I doubt this would ever be solved.

Disadvantages of the current setup:

• It's Dragon (which, as ethana suggested, "doesn't count").  The speaker was forced to run Windows in a virtual machine in order to run the program - it doesn't get much more proprietary than that.  Even the speaker's customizations don't seem to be available as of yet.

• Very expensive - he recommends a $300 microphone, among other things.

Maybe I was being overly optimistic in suggesting voice recognition would become more common - certainly not fast enough to get economies of scale.  The real limiting factor is the learning curve.  We all know how badly humans react to that. I'm guessing this would be last resort even for people suffering with RSI.

What I do suspect would happen is that we'd get language-creep from lobbying VR users.  We already see similar lexicon in paredit's "slurp" and "barf", so it's not out of the question.  Slowly, the barriers to entry would erode as monosyllabic words become more and more standard.  Though, by then, we'd probably have thought-recognition and all this would be irrelevant.  The median programming tool moves at a glacial pace.

I wouldn't think so.  Touch-typing is rather straightforward from the home-row position, and the rest is remembering some 26 keys you're probably already familiar with.  But, not having tried voice recognition yet, I don't know for sure.