That's a rule that might hold for applications and services. It does not hold for languages and libraries, where any and every aspect is going to be the bottlen

That's a rule that might hold for applications and services. It does not hold for languages and libraries, where any and every aspect is going to be the bottleneck in someone else's code. It's a different 10% for each user.