We were experiencing this for a long time, now we had some time to take a deeper look at the issue.
Sometimes the tracer session gets stuck. By this I mean that despite you have closed the tracer window, the system thinks your requestor is still tracing. Usually, a terminate trace or a logoff login solves this irritating issue.
Today we have experienced a this on a higher level:
The tracer cannot be started because an other requestor traces it, but that requestor does not even exist.
We know that it does not exist because it is not present on the requestor list, and we got the following error while trying to kick it programmatically: