But Reiser4, which never got merged, was insane levels of amazing in latency and ops/second, and yet excellent in throughput still.
Reiser4 was a team effort, and the other participants have been maintaining it irrespective of Reiser's imprisonment.
Today, there are some legitimately non-crap filesystem efforts, but they are either not mainlined and not very active (Tux3) or not Linux (Hammer2 from Dragonfly).
XFS could be made less crap; The main complaint I remember from Hans (dentries shouldn't be packed with inodes) does still stand. Metadata work such as a simple ls on a non-cached directory taking forever and a day.
I am most hopeful that when Matt Dillon declares Hammer2 done, porting efforts will make it widespread in open source operating systems.