The speed of input is not, I believe, what is most meaningful about using voice to deliver computer instructions. Rather, it is the voice's unique capability to convey meaning to other humans that text simply cannot convey. There is a power held within the human voice that has yet to be fully exploited by software, although Baker did much great work towards this goal.

Speech to text, to me, is so attractive because it may help to usher in a whole new, even more powerful input paradigm than anything else in-use today. It's not about speed - it's about precision.

EDIT: A study demonstrating my point can be found here: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3243272/