This is a classic example of a peer-reviewed article spreading a belief based on assumptions rather than measurements. Using a group result of varied usage patterns adds random noise to the data, so that the few percent of gained efficiency are likely not notable. Yet if you were to measure a single unit there will be effects.
How can I say that? As warm spring days came a few years ago I noticed the compressor of our rather new energy efficient fridge seemed to run more often than usual. Well, the room was warmer, I shrugged it off. One fine day, I noticed the ice cream I took out of the freezer compartment was rather soft. Got the thermometer, waited a while and read only a few degrees below 0, not the normal -18°C. So I cleaned the coils. The freezer returned to normal. I did not monitor the compressor or the energy consumption. Uploaded photo. Cocreatr 03:33, 4 June 2011 (PDT)