You then assume your own conclusion when you ignore pipelining. If instructions are executed in sequence indian-file then necessarily none will be faster than a

You then assume your own conclusion when you ignore pipelining. If instructions are executed in sequence indian-file then necessarily none will be faster than any other. The traditional rule-of-thumb is that programs have an ILP of two. The Execution talk (millcomputing.com/docs/execution) explains how the Mill turns that into an ILP of six. Then for the 80% or so of code that is in loops, pipelining has unbounded ILP - there will be a talk on pipelines upcoming.