The definition of a system's performance depends on its boundary. A car's performance is measured in miles per hour because the boundary of the "car system" is between the tire and the road. We could say that the car actually stops at the axle and that the wheels are a separate system. Then the performance of the "car system" would be measured in revolutions per minute. However, car and the wheels are generally considered the system. On the other hand, when we talk about a highway at rush hour the cars are considered subsystems of the traffic jam. Alternatively, an company's organizational chart is an illustration of subsystems within systems.
People who are in charge of a subsystem will generally consider themselves in charge of a system. When they strive to do the best job possible they will usually try to optimize the performance of their system. Just like the performance of the axles in a car is measured differently than the performance of the tires, the performance of one group is measured differently from the performance of the larger group it is a part of. The person in charge of the subsystem can't measure the performance of the system, because that's not where they are; all they have to work with is the performance of their subsystem.
This is a problem because to optimize the performance of a system you must de-optimize the performance of all the subsystems.
For example, a "tuned" car doesn't have the most powerful engine because it would rip the transmission apart. If the transmission were beefed up it would spin the tires instead of moving forward. If the tires were stickier it would warp the frame. If the frame were reinforced it wouldn't leave enough space for the big engine, or it would weigh too much and it would need a bigger engine, starting the cycle over again.
The Tsar tank. More like the reTSARded tank! Am I right?
A system must have subsystems that are in balance with each other based on the performance goals of the system, not on the performance goals of the individual subsystems. This is relatively easy to understand when the systems are not people. But as soon as people get involved they start to get all pissy about being a subsystem rather than a system.
This is why the executives of companies are constantly being reminded, often by highly overpaid consultants, that they have to explain to employees how their actions affect the company's overall goals. Otherwise, all they have to go on is the performance of the system they are aware of, which is the one they happen to be in charge of. When they do their best they will actually be destabilizing the company.
BTW, this is why companies alternately claim it is better to keep their employees powerless and scared, or empowered and brave, depending on which extreme they are already closer to. A company that judges its employees on how well they aid the overall goals will strengthen the company by empowering everyone. A company that judges its employees on how well they perform on their section's individual metrics will strengthen the company by squashing everyone.