According to Herb Peyerl:
>
>3500 I/O's to do a mkdir?
>

That does sound insane.  On my lowly lapdog with an IDE disk using ufs
I get:

[blymn@rover] csh
% time mkdir foo
0.0u 0.0s 0:00.03 33.3% 0+0k 2+6io 0pf+0w
% exit
[blymn@rover] uname -sr
NetBSD 1.5_ALPHA2

I know it's a lot of dicking around but you may be able to narrow
things down by:

1) splitting the mirror and checking unmirrored performance
2) using ufs instead of lfs

at the moment you have a lot of things that could be interacting, you
need to eliminate some of the possible causes.  If you can say, for
example, that LFS on raidframe raid1 sucks but ufs on raidframe raid1
is ok then this may help point the finger in the right direction.

-- 
===============================================================================
Brett Lymn, Computer Systems Administrator, BAE SYSTEMS
===============================================================================