As far as i know - you have to sent the downtime API request to the host the machine is monitored on.
But this can also be a bug or a design problem. I think someone has already tested this with a enterprise edition to say that it is also the same way there or that it is working. If it working inside the enterprise then it is a bug/limitation of the raw.
But i cannot say at the moment if it is this way. Normally it should be possible over the master as most times the single slaves are not reachable for every user.