A big part of this is that typing and dictation are very different from conversation. Conversation is free flow, very flexible to errors (since people can infer each other's true meaning) and interactive. However, in both typing and dictating, your end product is often meant to be read and so must be error free.

For dictating this means planning ahead to make sure that the sentence is well-formed. Have you watched yourself type? There's a lot of backtracking, erases and restarts. The feedback from dictating is also slower, which then affects your input rate. If you do manage to ignore this and speak free flow, even with perfect recognition (which is impossible even for humans), you will have to go back and edit grammatical and cohesion errors.

The advantages of typing are faster feedback (this is unavoidable as the computational needs will always be less), working memory augmentation and better nonlinearity for edits.

When you take into account all that, the time to correct grammatical and spelling errors, to edit for comprehensibility, I doubt that speaking will come out ahead because input bandwidth is only the bottleneck for trivial pieces.