When you computing something with Golem, your task is splitted into smaller pieces called subtasks. They are send to other nodes, which compute them independent (what makes computations more faster) and send you back the results. When you get all subtasks results, Golem will merge them into final result of your task.
you can define on how many pieces your task should be splitted. Splitting task to more subtasks allow compute it faster, because more nodes can compute them parallel. But the the more subtasks you select, the more it will take to merge the results, so it is not a good idea to split small task into many subtasks.
It's a good pracitse to try to split your task into pieces that should take ~20-30 minutes to compute, so if you are rendering aninmation with 50 frames and each frames takes around 10 minutes to render, you should split your task into 25 subtasks. If each frame takes ~ 1 hour to render, you should consider splitting your task into 100 subtasks.