I'm interested to know to what extent users and organisations are using FME Server as an Integration Platform/Environment to both Consume and Expose API Endpoints

A short reply based on my own experiences:

• For asynchronous requests (e.g. clip-and-ship) where there is no expectation by the consumer for an immediate result, FME Server is perfect.
• If the requirement is that FME Server reply to e.g. bursts of REST calls to return immediate results to something like a web app, FME Server may not be ideal for all cases. There are some ways around handling bursts, but it can be costly/complicated to scale for high volumes (reserved engines). This may get better with dynamic engines, however, but cost is still going to be a consideration for most cases if demand isn't fairly predictable.

I've had good experience putting up API gateways in front of FME Server to provide a customized REST API, thereby hiding the specifics of FME Server in the background, in particular for outwards-facing servers where, for security reasons, you'll want to hide as much of the actual infrastructure as possible.