I ran a rather simple benchmark over the past week or so to test the difference
between running a stock NetBSD system, and running the same system recompiled
fully (userland and kernel) with -march=i686.  These are the results.  First,
the stock compile:

kernel build of GENERIC (2 runs):
     3170.80 real      2792.83 user       240.49 sys
     3076.64 real      2794.78 user       237.89 sys
build.sh USETOOLS=never -T /usr/src/tooldir.NetBSD-3.0-i386 distribution:
    24578.18 real     19786.50 user      3336.71 sys

Now for the i686 built system:

kernel build of GENERIC (2 runs):
     3047.63 real      2628.01 user       242.87 sys
     2911.47 real      2623.92 user       244.22 sys
build.sh USETOOLS=never -T /usr/src/tooldir.NetBSD-3.0-i386 distribution:
    23951.30 real     19195.13 user      3348.42 sys

By my rough calclulations, thats about a 4-5% speedup on building the kernel,
and about a 2% speedup on building the base system.

Note that for all the timed builds the system was built with the standard
optimizations.  (ie, to not skew the results trying to optimize on one build
and not on the other)

These tests were done on a:
NetBSD 3.99.15 (GENERIC) #1: Tue Jan 24 21:21:11 MST 2006
        root@hadar:/usr/obj/sys/arch/i386/compile/GENERIC
total memory = 191 MB
avail memory = 179 MB
cpu0: Intel Pentium II (686-class), 397.96 MHz, id 0x652
cpu0: features 183f9ff<FPU,VME,DE,PSE,TSC,MSR,PAE,MCE,CX8,SEP,MTRR>
cpu0: features 183f9ff<PGE,MCA,CMOV,PAT,PSE36,MMX>
cpu0: features 183f9ff<FXSR>
cpu0: I-cache 16 KB 32B/line 4-way, D-cache 16 KB 32B/line 4-way
cpu0: L2 cache 512 KB 32B/line 4-way
cpu0: ITLB 32 4 KB entries 4-way, 2 4 MB entries fully associative
cpu0: DTLB 64 4 KB entries 4-way, 8 4 MB entries 4-way
cpu0: 32 page colors

With -current checked out around 1/23.

---
Tim Rightnour <root@garbled.net>
NetBSD: Free multi-architecture OS http://www.netbsd.org/
Genecys: Open Source 3D MMORPG: http://www.genecys.org/