Tape drive fails on read consistently

[Carl Miller <chaz@energoncube.net>]

I'm not really sure where to ask this question, and I'm at my wit's end,
so I'll try here.  I'll be happy to hear suggestions of where might be
more appropriate/constructive to post, help on how to debug this issue
further, or any other pointers.  And cc me directly on replies, please.

I recently got a new tape drive for my Debian server, as the backup sets
were outgrowing the old drive.  The new hardware is a Seagate Scorpion
240 autoloader.  /proc/scsi/scsi reports:

Host: scsi1 Channel: 00 Id: 15 Lun: 00
  Vendor: SEAGATE  Model: DAT    06241-XXX Rev: 8240
  Type:   Sequential-Access                ANSI SCSI revision: 03
Host: scsi1 Channel: 00 Id: 15 Lun: 01
  Vendor: SEAGATE  Model: DAT    06241-XXX Rev: 8240
  Type:   Medium Changer                   ANSI SCSI revision: 03

I can backup just fine.  So far, I've made four multi-tape backup
sets with this drive, and all has gone smoothly.  But I can't
restore.  Every time I try to restore a backup set, the restore
errors out 3/4 of the way through the first tape of the set with:

Failed to read /dev/nst0: Input/output error.

(A read() system call to the tape device is returning EIO.)  This
happens on every backup set, and always 3/4 of the way through the
first tape of the set.  (The autoloader has a neat little bar-graph
display of tape position, and it always stops at exactly the same bar).

I tried modifying my custom tape-handling utility to get the tape status
immediately after the failed read.  What I get each time is:

   type   word is 0x00000072
   status word is 0x05010000
   resid  word is 0x00000000
   dsreg  word is 0x25000200
   erreg  word is 0x00000000
   fileno word is 0x00000000
   blkno  word is 0xffffffff

(Those are the fields of the mtget structure returned by the status
inquiry.)  I'm at a loss as to what that might indicate, however, or
how to proceed in getting this resolved.  Any ideas?

Thanks,

                            -----Carl