From my experience working with it: in short, it is just a fancy name for shared memory between machines, with software libraries on top to manage synchronization, distributed transactions, et cetera.
From a sysadmin perspective it behaves less like a "single system image" and more like multiple images with shared resources.
My understanding that this clustering is more to handle external factors. It allows mainframes to be placed in different buildings anywhere from the same campus to up to 200kms away ideally providing enough distance to deal with natural disasters.
Mainframe itself (the "fridge") is a distributed system, as well as human brain is. It's all the management tooling that makes it appear as a single system.
Good luck trying to get five 9s when you're limited to a single DC. Cross-DC SYSPLEXing (GDPS) got us to five 9s and beyond without breaking a sweat. Our systems can never go down, ever.
From a sysadmin perspective it behaves less like a "single system image" and more like multiple images with shared resources.