Trouble reading dump tapes

[Jim Haynes]

The scene:  We have a number of VAX 750s and a couple of 780s, all
running 4.3BSD, all on the same Ethernet.  Some have tape drives,
Kennedy 9400 on Emulex TC7000 couplers.  Some of the machines
have Interlan and some 3Com Ethernet controllers.

Just recently I discovered in trying to recover files on tapes
made by rdump (at 6250 bpi, so block size is 65536 bytes), that
there are problems.  Specifically, the first reel of the dump
restores OK, but beginning with the second reel it will typically
say that it's doing byte swapping (which shouldn't happen) and
then blows up with lots of checksum errors and the like.

I haven't gone through all the tape sets in the tape room, but
  All the ones I have looked at made on the local tape drive, using
	dump rather than rdump, have been OK.
  Some of the ones made using rdump across the net are OK.
  Some of the ones made using rdump are corrupt.

I've pulled the original binary versions of dump and rdump and restore
and rrestore off the 4.3 distribution tape, to be sure that no local
hacks or library changes are involved.  Same results on the same tapes.

I wrote a little program to search for the magic number on dump tapes.
It gives results like this:
block 1   magic number at byte 24
block 16  magic number at byte 24
block 32  magic number at byte 1048
block 48  magic number at byte 2072
block 64  magic number at byte 3120
etc., where "block" is a 65536-byte buffer and "byte" is relative to
the start of the buffer.
Since I don't understand the dump program well enough to know where
the magic numbers should be I don't know if this is good or bad, but 
it's curious.
The /etc/rmt program is original issue.  The kernel has various hacks,
and in particular the Van Jacobson mods have been installed.

Any clues, anybody?  Is anybody else having this problem?