Any given CPU instruction potentially isn't in cache.

And yet, they can't all not be in cache. As soon as the cache is accessed for one instruction, a bunch more are pulled in too.

Any given vector.push call could reallocate the vector, but it's amortized O(1).