But the claim is that he is dispelling common falsehoods. Are there actually people out there saying that python glue with c libraries for performance critical parts is going to be orders of magnitude slower than an entire thing written in c?
However, if the experiment in question is, "using python glue code, can a developer string together a service that replicates the functionality of redis", then that illustrates what would actually happen in the real world.
For example, for a very junior Python developer, writing something in C is basically like asking a C developer to write assembly- possible, but they'd have to learn a lot of different things to do it.
Similarly, when I write Java, I want to be calling Arrays.sort(myarray) instead of writing my own sort (if the focus is on developer convenience and development velocity).
So, it's just a difference in what you're desiring to measure.