Going from 30 servers to 2 sounds like something whose cost savings might add up to as much as 100 hours of developer time.

That's cool. I bet they could reduce it to 0.15 servers by rewriting it in C.

They probably would have drastically reduced the number of servers even if the rewrite was still in ruby, having much better understanding around the runtime requirements.

Ruby also seems like a questionable choice in the first place for such a service. Perhaps a prototype.

Nobody is pretending that MRI Ruby is the most performant runtime you can find, but that's just not an issue for 99% of web stacks.