In a nutshell, MR tends to run through all of your data in iterations, which takes time and duplicates work.