I don't know how serverless platforms deal with it, but startup times could be "solved" by starting up the function and cloning it when it's ready to process requests, as Unix daemons have done for a long time.