The magnetron itself has about about 65% efficiency, but the paper conjectures that the longer duration of the pulses is due to defects in the cavity that result in some emission at a lower frequency (1.4 rather than the normal 2.4 GHz), so the energy radiated must be a tiny fraction of the nominal power.