Child particles can help give the appearance of many particles without slowing things down too much, which is especially useful when the particles will be combed or simulated.
But what about when using an object or group for a particle system which is not going to be combed or simulated? Are there any advantages to using child particles in this case?
There is very little advantage. Child Particles make copies of existing particles to reduce the computational power needed. When using objects as the display type for the particles, it is displaying a (translated) copy of that object at each and every particle. As far as render time goes, this will make no difference. When first setting up the particle system, using child particles will let you simulate the particles faster - but once you have the system set how you want it, it does not make much of a difference in render time.
Best way to answer this is to test it: The following data was gathered from a test scene with a particle system on a simple plane, and using default (non-manifold) Suzannes as the display object. All children were identical:
Suzannes: Actual Particles: Children Particles: Total Render Time:
10,000 10,000 0 00:28.19
10,000 1,000 10 00:17.02
10,000 100 100 00:21.85
Clearly, it takes the longest for Blender to render the 10,000 separate particles, as it must calculate all 10,000. child particles are less intensive, but only up to a point. When you get too many child particles, there begins to be no computational gain between calculating extra particles, and calculating a large number of duplicate particles. This is why one can expect the best performance by using some child particles (but not going overkill).
