Some additional thoughts:

In general, I  would agree to minimize the number of systems, but ..
with adsm it's little different: Some shops have definded downtime
windows for maintenance (HW and SW). These times are usually at night
or weekend. This may collide with your backup window and may force you
to decide between reducing service downtimes, buy another system or
think about your backup strategy.

In addition, IBM could not deliver (maybe it's different with OS/390
2.7) a fast tcpip stack, so the performance of those scalable systems
is limited in ip appls. I can install a ip stack for each ip escon
connection and a adsm for each ip stack, but then I have multiple adsm
servers on a single cpu, and that does not minimize administration. So,
in my opinion, adsm is not very scalable on mvs compared with other
applications.

And- I expect the difference of costs may come from hw and sw, too. I
don't expect to pay a million $ for a high end NT box, but a 9672-R26
is not far away (without disks!). Tape drives and libraries are
expected to be same price for all platforms. (Don't tell me you have
some for mvs, your ADSM guy gets them and don't give them back. So you
have to buy new ones for batch processing and SMS) For a M$, I can pay
2 guys for 3 yrs. If you upgrade a given S390, you have additional
costs for sw. You pay your cics and ims and db2 for all cpus in the
box, not just the used ones. And ask your mvs guys what they think
about an additional S390, and ask your Unix and NT guys and compare the
answers.

Scaling an adsm box is not as easy as saying "I have a mvs box, let's
use it." I payed for it, and I think others payed, too.

Roger