I think the issue here is that you're viewing this as too all-or-nothing. If the OS/2 subway software is well-tested and does everything it needs to do, what's the benefit of spending $10 million rewriting it or paying a guy to keep up with platform changes? If IBM (or whoever owns it now) keeps maintaining OS/2 so the software can run with little to no modifications on new hardware (since hardware fails), good on them. There's probably 1,000 such systems, and that's $10 billion in savings for society.
That doesn't mean no software should actively be maintained, a lot of it should. It also doesn't mean that working old systems shouldn't sometimes be replaced or upgraded. But it does mean backwards compatibility is important and shouldn't be dismissed.
The theory that a platform as 'backwards compatibility' is only good for lazy developers and lost source code. And yes, not doing maintenance saves time, but so does not upgrading your platform. I mean, if 8 inch floppies work for nuclear installations, OS/2 works for subway passenger systems, and Windows 3.1 for radio servers and other flight control systems, might as well not maintain anything at all. Heck, why not use relay logic in the subway control rooms.