Interesting question.
My immediate theory (and therefore almost definitely wrong) would be that we learn specific patterns for words/sounds in a language when learning to type (with two hands) as opposed to learning to write where every letter is written in sequence. So you "hear" the sentence in your mind and your fingers automatically translate to the most common pattern, much like the T9 dictionary does in phones.
I don't think there is a word for this phenomenon around yet, but the errors it produces are definitely homophone errors.
