> 
> 	I seem to recall someone commenting that -mpentiumpro gives a
> 	slight drop in performance for 386 and 486 chips, and a bigger
> 	gain everywhere else.

it was me probably. i've don some tests. -mpentiumpro looks almost as same
speed on i386, up to 5% slowdown on i486, but it's 5-10% smaller - takes
less memory, and finally faster on pentium and pentiumpro.
 
> 	If so it might still be worth considering for the default.
> 
> 	Actually, does anyone have a reference to a table of relative
> 	gain/loss and compatability for i386 -march and -mcpu options?
> 	It would be good to get that information up on the netbsd site
> 	to give more people the right information to select optimisation.

i could do some tests on many different programs... i have pentium and
pentium III.

----------------------------
Microsoft is not an answer.
Microsoft is the question.
The answer is "NO".