Its not a stupid question.

Normally when I have run latency calculations in the past I run them from the perspective of the caller, not the server.

In most cases this is over the network, a named pipe or sock file.

I guess it should be possible to run multiple runtimes inside a program that run independently.