md5 giving inconsistent results

[David J. Weller-Fahy]

I just installed 5.2.1-RELEASE on a RAID1 setup (atacontrol using
Sil0680 chipset), and was in the process of transfering files from my
old server.  However, I'm having an issue with the md5 sum of a file.
Here's my uname -a:

FreeBSD tigger.weller-fahy.com 5.2.1-RELEASE-p11
FreeBSD 5.2.1-RELEASE-p11 #0: Thu Oct 14 00:17:32 CEST 2004
root at tigger.weller-fahy.com:/usr/obj/usr/src/sys/GENERIC  i386

Here's the program I'm using:

> which md5
/sbin/md5

Here's the directory listing of the file (sized match on old server,
windows box, and new server):

> ls -al
total 714570
drwxr-xr-x   4 dave  dave        512 Oct 13 20:45 ./
drwxr-xr-x  16 dave  dave       1024 Oct 13 21:14 ../
-rw-r--r--   1 dave  dave  731327140 Oct 11 21:52 files.tar.gz
-rw-r--r--   1 dave  dave        162 Oct 13 20:44 files.tar.gz.md5
drwxr-xr-x   2 dave  dave        512 Oct 11 22:21 php/
drwxr-xr-x   3 dave  dave       1536 Oct 11 21:39 qmail/

Here's the md5 from the old server:

dave[tigger]/files> md5 files.tar.gz
MD5 (files.tar.gz) = 986a79245120f645b2a546fa6567d8ab

Here's three runs of md5 from the new server:

> md5 files.tar.gz
MD5 (files.tar.gz) = c57f4d82f32d698c16f0d792761ce9d6
> md5 files.tar.gz
MD5 (files.tar.gz) = 933608fc706c652b7d3ef2f5115d3ead
> md5 files.tar.gz
MD5 (files.tar.gz) = d8cd06f700b612fee72c39d5bc39df5a

And the md5 on a windows box:

C:\Documents and Settings\dave\Desktop>md5sum files.tar.gz
986a79245120f645b2a546fa6567d8ab *files.tar.gz

Just tried the same thing using md5deep from ports on the old and new
server, and got similar results: md5 sums are inconsistent when
calculated on the new server.

So.  What's my next step in figuring this out?  I was using the
following flags for optimization, but stopped and rebuilt world with no
optimization flags (except MAKE_IDEA=yes and LEAPSECONDS=YES).

CPUTYPE?=athlon-xp
CFLAGS= -O -pipe
CXXFLAGS+= -fmemoize-lookups -fsave-memoized
MAKE_SHELL?=sh
COPTFLAGS= -O -pipe
MAKE_IDEA=YES	# IDEA (128 bit symmetric encryption)
LEAPSECONDS=YES

I'm at a bit of a loss, and am about ready to try a reinstall, any
suggestions other than that and memtest86 (which I'll be trying next)?
Oh, and any pointers towards information that I failed to provide will
be gratefully accepted.

Regards,
-- 
dave [ please don't CC me ]