For z/OS, I don't think much anymore. The high level of single-machine redundancy and architecture of I/O offload USED to be better than commodity machines. That's been eclipsed by raw power and improved distributed functionality in the commodity world.

There are still companies (Visa, Airlines, GDS) having issues getting off the more unique TPF operating system. It's basically a distributed K/V data store that scales very well and deals well with write contention. So far, only Amadeus has been able to migrate off of it. Some smaller airlines were able to migrate away, but not the larger ones.

The company that Google bought (ITA), was on their way to writing a full reservations system. They had finished the shopping engine but not the whole system. That was a company chock full of money of talent, and they didn't finish. Though their shopping engine is the best one that exists.

Edit: Somewhat related point. The "developer efficiency" was quite good in these environments. Since the OS came with everything a developer needed, there was mostly only one right way to do everything. The languages, libraries, database, batch management, "UI", authentication/authorization, security, and so on, was already decided for you.

Think: opposite of general-purpose operating systems use-cases. Very IO/compute/latency-intensive things. Where the whole programming paradigms are designed around efficient (even real-time) compute. Where the software+hardware pair are written once and rarely if ever changed, restarted, or taken down for maintenance. Credit card transaction processing and airline reservation systems for example.

Of course you could use Linux or similar for such jobs, but highly specialized hardware+software pairings can eke out surprising performance at the cost of massive hardware and software-engineering bills.

Some software was written for Mainframes that effectively NEVER reboots. So you have to modify it to support reboots easily, but it was written in COBOL or and old version of C and compiled with an old compiler. So it isn't that easy to fix.

My first professional dev job was COBOL and it's associated tech, a majority of the stuff we did was banking stuff and at the time it seemed like that's what a majority of the COBOL market took up but I could be wrong