Yep.  We send it on every work change request that isn't difficulty adjustment (which we send again after 30 seconds with the flag if the user isn't respecting the new difficulty).

But we're seeing a TON of shares submitted a significant amount of time after the clean jobs request.  A lot of them 15-20 seconds after telling them the old work isn't valid, and those shares are still being accepted, which leads me to the guess that people have pushed their miners to max out hashrate (and in this case accepted shares) rather than actual accepted shares.