We have (had) similar problems after upgrading to dlt7000 (from dlt4000), for
several months. Two weeks ago we changed (not the first time!) microcode
in dlt7000, another fix in adsm etc etc etc. Now we had 2 weeks without
i/o error (hear, hear). On my stk silo connected to s390, I never had comparable
problems.

Some days ago someone told me that quantum (only manufacturer of dlt drives) 
says
that those drives like to have a constant datastream of at least 3.5 MB/s, 
otherwise
they have to reposition various times which results in lower reliability (I 
didn't check
this but it sounds good). We are backing up through the net direct to tape
so we can't reach this throughput.

Maybe the reliability is improved when one is using disk pools as cache?
Or watch for throughput of the affected scsi channel (or both)?
Or the adsm uses smaller tape buffers than other comparable products do?

I don't think this problem lies only on stk side. Now we have additional adsm 
servers
with 3590, you may ask me middle next year if it's better.

Regards
Roger