On the other hand the program you are piping it to likely needs to do some setup as well. Plus with the fact that shells probably spawn the pipeline in order an

On the other hand the program you are piping it to likely needs to do some setup as well. Plus with the fact that shells probably spawn the pipeline in order and that the setting up the buffer is incredibly cheap it would be incredibly rare that the first write wasn't done before the program did the first read. ...of course this can be more complicated if you are CPU starved and these programs are competing. Probably the more useful metric is total efficiency. How many interactions need to occur before the setup cost is made back up. And how many lines are read on average from yes?