If you detect 50% errors each time, it doesn't matter how many times you edit, you can't be sure you've found all errors. Add a couple factors such as tiredness and a small but real possibility you make a mistake while correcting and error, and you could spend hours on a letter.
This is where you introduce a concept of results against effort. When you first edit, there's enough mistakes that correcting them is worth the effort, but at some point the time spent rereading will outweigh the gains. You've gotta stop there and leave the text as is.
(Note that the result vs effort ratio can be heavily affected by used tools such as spell-checkers.)