Very interesting question, thank you very much for your feedback.
What I try to say here is: 50% of all rewards go to posts below $6, and 50% go to posts above $6.
Then $6 is this middle point. It corresponds to the median if we define the population as the number of rshares (not the number of posts) in a discrete probability distribution (wiki ref).
Also you can see here https://joticajulian.github.io/steemexplorer/#/reports this distribution. I draw the median with a different color.
This value is very related with the previous hardfork, where we had linear rewards: Posts above $6 get more rewards compared with HF20, and posts below $6 get less rewards compared with HF20 (not taking in consideration the cut of 10% for sps).
Thanks for the clarification.
Indeed, the curve does what you'd expect it to do: distribute the rewards more unevenly among posts. The rationale for that is that is makes it easier to limit the amount of value sucked out from the system by spam and farmed low-quality, low-value posts flying under the radar. Many people expected this to result in worse non-merit-based income inequality than before. But increased downvotes put an end to the bid bot industry, which made former bots curators, which in turn made the whole platform much more meritocratic.