Optimize network thread's batching overhead slightly - during the end of frame flush only set the last batch sent as "immediate" priority - avoids paying the extra 0.4us to wake the RakNet thread after every single batch sent, instead only paying it once for the entire batch