Some rules of thumb for OpenBSD optimizing:

   1) Don't try to optimize things if there are no real performance
      problems or threats of such problems.

   2) Don't guess. Find or run benchmarks.

   3) Remember that benchmarks are not real applications.
      Only way to find out if a real application does its
      job properly is to run it in a production environment
      for at least a month or two. Sorry.

   4) Benchmarks are often specific on specific hardware-software
      combinations.
 
   5) GCC optimizations, such as recompiling the system with
      -O3 -march=i686 instead of letting it go at the default
      -O2 [-march=i386 implied for x86]) are usually a waste of 
      time. There are, of course, exceptions, but they rarely
      are system-wide but are more closely related to one
      math-intensive library or application.

      Debugging and profiling compilations don't usually 
      significantly slow down programs, but they offer the wonderful 
      opportunity to find *real* bottlenecks.

   6) Aim for future scalability.

   7) Performance doesn't matter if you don't also have reliability.
      500 hits per second doesn't help if you only get 12 hours a day.

   8) Remember to read your configuration manuals carefully.

   9) After you've read the configuration manuals really, really
      carefully, re-read them, and pay some attention this time.

   10) Calculate one hour of your work configuring, compiling, 
       optimizing, aso. costs to your company or organization, and 
       compare it to updating your hardware or outsourcing the thing.

Things to read:

   http://www.openbsd.org/faq/faq11.html
   man sa(8), x11perf(1), systat(1), vmstat(8), top(1), ps(1), 
      iostat(8)
   http://www.kegel.com/c10k.html
   http://perl.apache.org/guide/performance.html
   http://httpd.apache.org/docs/misc/perf-bsd44.html
   http://httpd.apache.org/docs/misc/perf-tuning.html
   http://www.ontek.com/mikey/optimization.html
   http://www.daemonnews.org/199908/portability.html
   http://www.spec.org/spec/faq/