The idea of a book on tuning Linux for performance is pretty much a moving target. Linux makes advances at a rapid pace, and things change often. Since its introduction in 1991, Linux has gone from a state where it could not compile itself (and barely boot) into a system that can take up to three full CD-ROMs to install much of the software. It rivals operating systems twice its age in terms of performance and stability. Its complete heritage may never be known because of the thousands of people around the world who have contributed patches, drivers, and enhancements to make Linux better.
This advancement means that some portions of Linux can literally turn on a dime. The Linux 2.0 kernel used a program called ipfwadm to administer its firewall settings. When 2.2 came out, the program was ipchains, and used completely different concepts than ipfwadm. The 2.4 kernel uses iptables, which itself is different from its predecessors. This means that Linux has little baggage holding it back when it comes to backwards compatability. Some open source applications need no changes, most may need to be recompiled, and some require programming changes that can be done by any programmer. The idea of backwards compatability in Linux is a foreign concept.
But tuning Linux is more than just tuning what you have. It is looking at your requirements, evaluating the technology that can solve that requirement, and going with the balance between cost and performance. Not everyone can afford the best cards, but the difference between good and poor performance can be a very small cost as compared to the overall system cost.