All of which is a case for delineating language content by TLD. It's easy enough, with fewer dialect subdirs than the monolithic techniques often employed by sites, but IME TLDs just offer a default with all the complications you describe carried from region to region.

But hey, Google's always been this way. Before you could even log in they used to have a "Settings" pane where you cuold set your language to be saved in a cookie. It would get lost all the time; you'd always have to reset it.