Also in the context of cloud spend: way way way less RAM, which can translate directly to dollars.

Yeah, I just wrote a simple daemon in C that had no performance requirements ( listening to a udp socket and dumping stuff into a pgsql db once a minute) because the Python program would use like at least 20x the RAM. When you're running a bunch of things on a resource constrained place (e.g. a single computer that has to do a ton of things sitting on a rack on the Greenland ice sheet), even just the base Python memory usage from a new process adds up...