If you elide such a feature
from the entire language then you also won't be able to build a fast (standard) library. One of the reasons you usually don't have to overly worry about performance is that the standard library (as well as other libraries) are typically pretty optimized. If you take away the possibility for that optimisation then the entire language becomes slower, by virtue of having a slower standard library.
"Artificial microbenchmarks" are very useful if there's a good chance the code might be called a few thousand times in a loop. If you write generic code that may get used by thousands or even millions of applications then these small differences do matter, and add up.
Also I think a lot of code will be needless awkward too; sometimes I really just want to get the first or second or last index. I don't want or need to iterate: I just want exactly nth entry, nothing more, nothing less. Yes, you need to be careful with it, but entirely removing them is not much of a solution.