Has anyone calculated the marginal Watt per increase in GH/s?

Say you clock from from chip_freq 393.75 (202GHs) to chip_freq 400  (205GHs).

The increase in GHs is 3. If this requires 12W more, then the marginal Watt/GH/s is  12/3 = 4. Hardly profitable (depends on $/kWh).

I think mining has reached the point where overclocking may no longer pay off. However, it needs to be tested.