Workaround library bug in eventloop(). Apparently if there is little
activity and we catch the timeout once and we continue, then if there
is some other error like server killed or otherwise connection lost
then it will refire as a timeout, and this happens endlessly, so
resulting in 100% CPU loop.
We now detect it by (far) too little execution time, ah well, it is
a workaround that works :D.
So this should be safe now (was introduced a previous commit,
still the same code):