An hour lost on the bottleneck is an hour lost on the entire system;
An hour gained on a non-bottleneck is a mirage’
There isn’t a more grave mistake than to equate non-constraint with non-important. On the contrary, due to the dependencies, if you ignore a non-constraint it can impact the constraint to the extent that the performance of the entire system severely deteriorates.

What is important to notice is that the prevailing notion that “more is better” is correct only for the constraints, but it is not correct for the vast majority of the system elements—the non-constraints. For the non-constraints, “more is better” is correct only up to a threshold, but above this threshold, more is worse.
This threshold is dictated by the interdependencies with the constraints and therefore it cannot be determined if you examine the non-constraint in isolation. For the non-constraints, local optimum is not equal to the global optima; more on the non-constraints does not necessarily translate to better performance of the system.
We now recognize that the vast majority of the elements of a system are non-constraints. We also recognize that for non-constraints more might not be better but worse. So, what must be the unavoidable result of following the prevailing notion that more is better?
The number one reason for not doing what should be done is doing what should not be done. We don’t have a choice but to define focus more narrowly: do what should be done AND don’t do what should not be done. Learn more about Theory of constraints based solution.