>I'm saying that things that don't fly at small scale, don't become good ideas at large scale.

They absolutely do.

Fourier transform multiplication methods are a terrible idea if you run them on numbers that aren't hundreds of digits long, but the only way to multiply truly large numbers.