IIRC this limit has existed since at least 2004, with no discussion of improvements in hardware and network administration tools in the meantime.
The use case would be something like an internal corporate knowledge management system. Page/revision tables would perhaps be large, but few queries would run at any given time. End users would furthermore be accustomed to waiting, say, 30-300 seconds for a page to be served (from their experience with bloatware desktop apps and commercial cloud portals). Every so often, a large history would need review: legal discovery, regulatory audit, technical debt remediation.
Obviously such a limit would never be raised from its default on WMF projects, with their immense volume of gnoming transactions, and be accompanied by a sulphur-and-brimstone comment in DefaultSettings.php about the risk of DoS attacks and such.