Posts don't have a language code in them so we need to download everything, process all of them of language code (which fails in 10% of the time) and then delete everything that doesn't match the user language. It's a lot of data usage. 

But the main issue is that it creates local bubbles. If you use Twitter for instance, you rarely see Arabic or Mandarin posts, even though there is as many of them as in portuguese. The client just hides all external viewpoints from you. Making users a slave of their own linguistic cultures.