Not to mention when the processing of huge set of data is not for having a short answer but to process the data into an other form, the same size or still big.

Not to mention when the processing of huge set of data is not for having a short answer but to process the data into an other form, the same size or still big. Even if the data is massively distributed from the start the communication and aggregation of distributed data between processing or client nodes might be a considerable bottleneck. Different aspects might desire different logic for distribution otherwise the performance suffers, but rearranging for the sake of processing might ruin the overall performance of the task.