As usual, taking the average of young faces makes for a pleasing effect. Much of facial attractiveness is lacking random defects, which creating composite digital images washes out. (On the other hand, while averaging can get you pretty close to, say, Britney Spears at age 20, it doesn't seem able to get you to, say, Audrey Hepburn.)
In commenting, please keep in mind that Roissy's post is a good place for Roissyesque comments, while this is a good place for discussing the various high-minded epistemological issues I will bring up below.
Unfortunately, I can't find the source of all these pictures. (I'd guess yearbook pictures. If so, we're dealing with young women from above the dropout class. Or maybe they are beauty contest participants. Maybe one country's photos are beauty contest winners and another country's are mug shots from county jail. I don't know.)
One thing that's interesting is the difference in facial expressions, which has a lot to do with social attitudes. For example, the composite West African girl looks sweet, while the composite African-American girl looks like the cast of Waiting to Exhale is telling her "You go, girl!" And the poor composite Afghan girl looks like she's kind of worried that the Taliban will at any moment break in and set the photography studio on fire.
On the other hand, a lot of differences in expressions may just have to do with what the individual photographers asked for, and other non-national semi-random characteristics.
For example, which is the German composite, which the Greek, and which the French? (This is not a trick question.)
I'm not sure why the middle face is shorter in the vertical dimension than the other two—this probably has more of a technical explanation involving how the pictures were taken or the images recopied than a physical anthropological explanation. I wouldn't be too excited about comparing the composites versus each other since that's so dependent upon how pictures from around the world were taken.
Instead, what I think is interesting is how much each composite is fairly recognizably traceable to relatively close to its country of origin.
It would be an interesting experiment to do on a large number of people to see how much more accurate their national identifications are when using the 41 averaged composite faces versus 41 individual faces from the underlying samples.
The existence of types is quite unpopular in the human sciences these days. For example, the Field Museum has a stupendous collection of 104 sculptures they commissioned from sculptress Malvina Hoffman in 1930 to illustrate "The Races of Mankind." In 1999s, I asked a Field Museum official why the museum didn't treat this amazing resource with more respect. She said they weren't realistic because they were "types."
In other words, when Hoffman sculpted, for example, her Nuer tribesman of the Southern Sudan, she sculpted an extremely Nuery-looking fellow imaginable—about 6'8" and 140 pounds. In reality, by no means is the average Nuer 6'-8". On the other hand, the Nuers, like their archrivals, the Dinkas, are among the most "elongated" people on Earth, so Hoffman definitely caught their particular tendency.