At high-load currents, the power dissipated by the load was much larger than the power dissipated by the resistive-feedback network. This is why the efficiencies for different R1 and R2 values converge at higher-load currents. However, at
low-load currents, the differences in efficiency for different feedback resistances are more prominent. This is
because the current through the divider dominated the
current through the load. Therefore, to have higher efficiencies at light loads, it is good design practice to use the
large feedback resistances recommended in the datasheet.
If efficiency at light loads is not important in a given design,
then smaller resistances can be used with essentially no
impact on efficiency.
Отсюда:
http://www.ti.com/lit/an/slyt469/slyt469.pdf