Is there a performance-critical case that isn't already ruled out by the Lambda cold boot problem though?

nor would you use a dockerized microservice for that, or would you?

serverless is - as i said before - a dockerized microservice as its heart. It should only be used in in places where you'd do it without the abstraction.

There are a lot of services / applications you can build with this. For example Adapters for external SaaS which should be able to trigger certain actions, or just plain JSON APIs which query a DB and output their results...

but using it to join 2 TB of data and process it afterwards in realtime? yeah, thats not a valid usecase for serverless.