The benchmark in that article comparing JWT to opaque session id's doesn't mention what database setup is used or what JWT signing scheme is used.
Presumably it's just using a HMAC, I think if it used RSA or an asymmetric signing scheme, the performance would look much different.