Are you worried at all about their patents stance. I currently I think it says if you litigate with Facebook you lose the license. Otherwise I agree zstd is looking like a very nice improvement in an area where most people think nothing happens. I especially dictionary compression bit.
That's a good point, the downside of using new algorithms is playing by the rules of their patent licenses. In this situation I'm not concerned about litigation with Facebook, but I could foresee other products where that might be an issue.
Aside: anywhere you store compressed data at rest, you need to store (in a header somewhere) the algorithm used to compress that data. If you need to change algorithms down the road--e.g. you do get into a lawsuit with Facebook--you'll need that header to know how to decompress old data vs. new.
