You can't compare optimizations done on audio encoding/decoding (often hand written assembly) to something like PostgreSQL which has many moving/inefficient parts.

The optimizations on PostgreSQL are real if done correctly, it's just a very big project.