Niagara performance - best practice and perf

Howdo, I’m looking for some performance information, or best practice for using Niagara systems and emitters.
I have hundreds of systems on screen at once and need to optimize. Lots of them use the same group of emitters with the only difference being position and a mesh normal value.

I’m trying to find if it is more performant to copy the group of emitters to lead particles, (like in the fire work example with a head and trail), but the emitters are more complicated than just trails and need to spawn from a spheres or boxes

…or if I make the emitters in to master emitters, are they kept in memory, like a material function or master material, and hence would be more performant,

What is the over head of a system or an emitter, at low amounts of systems it seems negligible, but at larger numbers it must take some cpu time.

Cheers