> (in my opinion, it will be quite hard for them to achieve <100ms startup time with their current architecture).

Who's running python workloads with sub-100ms latency requirements?