The measurement itself will affect the event queue as well but this should have a much lower overhead than a full profiler.
This also takes cpu intensive events into account (which would not be possible by just looking at the # of events).
While I wasn't able to find the number of waiting events in the queue I found another health metric that might be useful:
delay will contain the milliseconds it took from queuing the event to executing it.
would be great if we could actually 'see' the event que... I suppose you never found a way to do this?