One thing that's interesting to me is how noisy (in the communications theory sense of the term) ordinary speech is. We all edit this out constantly, but if you look at an accurate transcription of almost any speech stream, it's full of garbage -- repetitions, dropped words, misused words, stammers, syntactically neutral placeholders, random noises, etc. I myself stammer on a fairly regular basis, but I don't think many people notice it.
Yet in written text, there is (rightly) a premium on clarity. Even written dialog, as within a work of fiction, can sustain only a small amount of that 'noise.' Consider how negatively some people react to "eye dialect" [ Wikipedia ]. I have to read portions of Tom Sawyer aloud, and all of Riddley Walker has to be read that way to be understood in the slightest.
So written text, even when it's aspiring to represent the spoken word, is offering a highly idealized version of speech. Likewise the potentially complex structures of written text correspond to the ordinary speech patterns of no one this side of William F. Buckley.
Kevin J. Anderson reportedly writes some of his books by dictating them into a recorder and having them transcribed. For me, the language which flows out of my fingers is so distinct from the language which comes out of my mouth that I might as well be working with two different brains.