Bizarre RAID "failure"

Hi, all,

I'm encountering a bizarre problem with software RAID 5 under Linux that
I'm hoping someone on this list can help me solve or at least
understand.

I've got a box running Red Hat 7.3 with SGI's 2.4.18 XFS 1.1 kernel. 
It's using three FastTrak TX 2000 (PDC20271) cards in non-RAID mode with
three Western Digital 200 GB drives.  I'm using those controllers
because they were handy and they support large drives.  The drives are
in an XFS-formatted RAID 5 array using md, which has never given me
problems before.  In this case, however, I'm running into some seriously
anomalous behavior.

If the system goes down unexpectedly (e.g., because of a power failure),
the RAID array comes back up dirty and begins to rebuild itself, which
is odd enough on its own.  What's worse is that, whenever this happens,
the rebuild hangs at about 2.4%.  When it reaches that point, the array
becomes totally nonresponsive--I can't even query its status with mdadm
or any other tool, although I can use "cat /proc/mdstat" to see the
status of the rebuild.  Any command that attempts to access the RAID
drive hangs.

My assumption would normally be that there's a hardware failure
somewhere, but I've swapped out each component individually (including
cables!) and the same problem keeps happening.

Has anyone seen this behavior before, and can you recommend a solution?

Thanks,

Tom