Low level tweaking for one CPU, on one operating system - manual tweaking 
(except in certain RT applications which may have very tight timing and 
memory constrains - is not worth it either.  Too little gain for tons of 
manual labor.
 
Any tweaking for normal(non-RT apps) should be done statisitically with a 
profiler.  This gives the best 'return on investment in programmers time.
 
Further, if a profiler is used in a multi-tasking environment, and 
improvements can be made in a particular task, then that is all that can be 
said.  To look at making improvements in how the overall system will do one 
needs a 'system profiler' that may run in the background for days or several 
weeks - depending on the mix of tasks that will run concurrently over that 
time period.  Now one must try to find the right mix of tasks running 
concurrently (production tasks) that will optimize the entire system - 
including all it's peripherals.
 
Even that is a questionable activity if hardware is getting cheaper almost 
weekly.  It is only when hardware is expensive that doing a system 
optimization is worth the labor cost involved.  Just my two cents.  Have a 
good one.
 
Regards, -= Lou =-
---
---------------