Every pull request can theoretically be processed independently. But we run all of them sequentially, consuming a lot longer than required.