I'm almost on the verge of advocating to keep the status quo as is, but I do see your point.
A few facts for context:
The Reviewer and Steward badges (for reviewing 250 and 1,000 tasks in a single queue, respectively) count towards moderator candidate scores in elections, meaning that the SE team considers a user who has one or both of these badges as more likely to be a better moderator than one who doesn't. In my opinion, steps should be taken to make sure that users can't cheat their way into getting these badges.
It's possible for really savvy users to figure out which tasks are audits and which aren't. This does require paying some attention, and the whole point of audits is to ensure that reviewers are paying attention, but it's possible for users to game the system. This gets even easier if the user uses one of many user scripts that others have written to detect review audits.
I disagree with the solution proposed in the question, because reviewing an audit usually takes as much time as reviewing a real task, and I feel that legitimate users should be rewarded for their effort.
The solution proposed in Rob's answer seems like a good idea, but it's easy to game, especially on sites with a large amount of review tasks in queues (such as many of the queues on Stack Overflow). If Rob's answer were to be implemented, one could do the following: review 19/39 real tasks, change their strategy to review only audits for a while (while skipping all the real reviews), then when no more audits show up after some time, just make another real review to finish off 20/40.
This means that that solution would make it easy for users to game badges that help differentiate moderator candidates in elections, and I believe that that's a very bad thing.
To that end, I'm putting forth an idea that would help solve all problems: audits should allow users to continue reviewing, legitimate users should be rewarded for passed audits, and users should not be able to game the system to earn those badges. This is a mixture of the solutions proposed in the question and in Rob's answer:
- Audits should not count towards the daily review limit.
- Passed review audits should count as a review done, up to a daily maximum number.
- After a certain number of passed audits per day, users should continue to be given audits (to check for attention), but passed audits shouldn't count towards the total number of reviews done.
The "maximum number" above should be high enough so that legitimate reviewers keep getting rewarded for their effort, but low enough so that users who try to game the system get stopped. I'd say, around 3-5.