One of the most important aspects of performance tuning
and even programming in general is knowing when to stop. Getting
your web site or product into the hands of the customer is after all
the goal of all this. If you sit around and fiddle with it for months,
nobody is using it.
Once you have located all the low-hanging fruit, further
optimization tends to get exponentially harder for less and less
gain. The time you spend on it compared to simply buying another
server or two is usually not worth it. Stop and move onto the next
project.
Or better yet, go home, spend some time away from these beastly
computers and spend some time with your wife and/or kids