... it's it's quite likely that the program that prints 1 header will be run 100,000 times.

Then why not profile the entire program, including network IO, to see if the cost of loading the module is even significant?

Is "it doesn't take all that long to load" really a good argument to advocate the use of a module?

The real question is "Without actually measuring, is worrying that a module might take too long to load sufficient justification to write my own minimal replacement?"