Overdraw with transculent materials are really expensive (redrawing the actual star and cloud texture information on screen).
The method that **@ ** have presented runs only once (calculates the particle positions when the particle system gets created) therefore it have no expense later. Shader Complexity of the projected content can always be inspected to see what are the real implications of alot of overdrawn pixels.
This c++ implementation is still a little less expensive than creating similar results purely with materials and not using this engine modification at all. The key difference here is that the galaxymap informations can be recalculated always at runtime, or you can simply generate them only once at the begining of the particle system’s life. But it have very little expense actually to recalculate them live. GPU’s are rather good at calculating stuff. 
A particle material is capable to modify the world position offset of each individual particle, which can be calculated at runtime:
Implemented the same calculus from the c++ code purely into materials using built in nodes and it works just fine
After a little tweaking and adding a some effects to the material the result looks even better
The real expenses here the transculent material, and more importantly the blended-textured areas, they generate the real expenses (eg the smoke)
Even, if your material looks this large, the calculus can have very little footprint on the actual peformance
