Plus even at AWS scale, I imagine HW homogeneity of the fleet is more important than efficiency of a single application, unless it's insanely more efficient for a top 10 service. Homogeneity improves predictability, load balancing, operations, security, and all sorts of other things.
On the other hand, AWS offers AMD machines and ARM machines in addition to the typical Intel stuff. If there was some new architecture that offered a 50% speedup, and all you had to do was recompile your code, I assume they'd support it in a heartbeat.
In the context of the article, the changes would by necessity be far more invasive than "recompile your code". The premise is that C cannot be an effective implementation language. A full rewrite would be required.
