One difference between the two is that a system process is aggregated, which means it contains a inventory of all elementary flows combined for the preceding life-cycle stages = life-cycle inventory (LCI). So there's no need for providers, it's all in there.
On the other hand unit processes contain product flows that can be and should be tracked back to other processes or providers.
Which means using system processes results in smaller product systems with less links and thus shorter computation time / less CPU / less RAM usage but you loose detail as you cannot track back where your impacts coming from.
Also, a database containing only system processes is usually 10-20x bigger in size than the same database with only unit processes.
