This is entirely explained by game theory. If you think the final star rating for something should be something other than what the average currently is, you maximize your power over the rating by using only 1 or 5 star ratings.
That's assuming reviewers only derive utility from some product's star rating and not from voicing their opinion. If it were solely game theory, we'd see tons of 1- and 5-star reviews with only one or two lines of text (as this would maximize utility while minimizing the time cost of reviewing). For websites with more affluent user bases, (Yelp, Amazon's non-video game products), that's not the case: there are some epic screeds for and against restaurants and coffee tables.
To me, it looks like a self-selection issue: people only review a product if it either (a) changes their life positively, or (b) causes them trouble. If a product only impacts someone's life in a trivial way, well, why write a 3-star review for a trivially good product?
This is entirely explained by game theory. If you think the final star rating for something should be something other than what the average currently is, you maximize your power over the rating by using only 1 or 5 star ratings.