Phil, we had almost an identical problem (the millions of records)
when UMSLVMA (or B, I forget which name it was at the time) switched
our link to run with RSCS V2 at their end.  We had a file (of about
1000 records) which would go up to them, and then turn around and
claim to be sending 13000000 (13 million) records back to us.  That
scared the daylights out of me, because our spool wouldn't hold
nearly that much, and it would tie up the link for hours, of course.
 
But it didn't!  It would finish in about 10 or 15 minutes, and then
start back up to UMSLVMB.  The best we could figure out was that
some sort of transmission error occurred at the end of the file, and
RSCS V2 was going crazy in its responses.  They went back to a V1
end (and we back to VMB) and everything worked.
 
While I'm not defending V2, the problem may well not be just a V2 one.
It may well be a problem with the way V2 and V1 NJI links talk to each
other.
 
Richard