As with all other "preferable" or "best" approaches, it depends on what your end goal is.
For example, if you associate the particle list with it's emitter, this means that the emitter must continue to live until the particles have all decayed away. So your emitter needs to have more state. It needs to know whether it should be actively emitting more particles or in a "wait until all particles are gone" state. The latter doesn't make new ones. If you don't have this multi-state system, then the particles will all vanish the moment the particle system gets rid of them.
"Global" particle systems (it doesn't have to be "global" in the strictest sense. Just not bound to an emitter) allow you to de-couple emitters from the particles they emit. So your emitters can be very simple. Even better, you now have the ability to do different things you couldn't before. For example, you might want particles to respond to forces, like having particles near an explosion be pushed away by the impulse. Doing that with an emitter-based approach would be painful; you'd have to iterate through every particle emitter to make it work. With a "global" approach, you don't.
Then, there's the question of depth sorting. If multiple particle emitters start to interact, how do they sort with one another? Obviously this requires some level of 3D for it to matter. Again, the "global" approach makes this easier. Of course, you may not care about depth sorting between particle emitters, particularly considering sorting cost. But you have the option of caring with the "global" approach.