Let's re-run Tilman's query (T149021#3287887), but instead of looking for auto-confirmed users, look for auto-patrolled users. There isn't an easy way to actually determine if a user was auto-patrolled at the time of article creation, so we'll have to use current user rights as a proxy indicator. (This should be a much simpler query than the one for auto-confirmed.)
This data, combined with the existing data should let us answer (with some caveats) the question of how many new articles (on average) go into the reviewing backlog per day.