Herbie,

 > before I changed to "Profiler" I was thinking about  
 > the same as you do now but things are rather strange....

Could you prove/illustrate the implicit claim that TICKCOUNT 
used to measure program speed over intervals of several 
seconds is unreliable? 

What about a stopwatch or a wall clock?!

It is quite conceivable that an oldfashioned benckmark of 
the sort I posted is gradually becoming meaningless because 
the current multitiered RAM structure gives them 
unrealistically fast RAM or just plain unpredictable RAM 
service. But, before swallowing such a bitter pill, I would 
like to see some hard evidence.

Have you encountered important speed differences between 
stack variables and global variables? 

Incidentally, some description of the modus operandi of
Profiler would be both helpful and confidence-inspiring.

Cheers

Laurent S