Very few system calls in the main loop. Not even close to being io bound, and as I said memory is all preallocated, so allocator efficiency isn't a factor.