I'd trust those benchmarks to be indicative of problems in ext4 as far as
I could throw their designer.
e.g. one really obvious case they haven't checked is whether other
filesystems and raw block devices are also affected. If they are, this is
due to changes in the block layer (of which there have been many more in
the target time period than changes to ext4), and Ted is blameless.
TBH their PostgreSQL results are indicative of a misconfiguration of some
kind more than anything else. Do you think no PostgreSQL users anywhere
would have noticed a *factor of four slowdown* between .31 and .32 and
mentioned it on the kernel list?