In some cases increasing/decreasing log level with runtime commands can lead to a complete freeze of the server.
The only way to recover from this is "kill -9"
This affects at least trapper process, but is probably not limited only to it.
It seems that this is affected by server size, as I have only experienced this on large installations (>10k NVPS)