Actually, it would call for a higher refund as the power efficiency would also be impacted by either 17% or 33% (in your example).

It would call for an additional 500$ refund if the power efficiency would be off by 17%. (calculated on a basis of 100$/kW/m hosting cost and maximum power draw of 2.7kW at the wall, over one year only)

This is a conservative estimate, but let´s wait until final specs are released.