This is a clever design without the need for an opto coupler to sense the secondary side and to feed the primary side to regulate the secondary output.Many people will like to use this design even though the output load regulation is average.
Not a very new technique, but much more practical when integrated into the IC. My only gripe with this method is the accuracy obtainable. I tend to do the sensing on the secondary and then use the opto as a critical switch. You get better regulation and the opto plays less of a role in accuracy. Stability is still an issue of course, but it becomes the same for each unit built making it something you can deal with.
Krh, to answer your question:
1) 85% peak efficiency for 5V output and 87% peak efficiency for 12V output. A more detailed discussion can be found (http://cds.linear.com/docs/LT%20Journal/LTJournal-V22N3-06-df-LT8300-MinChen.pdf).
2) It does scale and you get better efficiency for higher output power. Check LTC's No-Opto Product Family.
@ETmax of course the ordinary crowd does it with an isolated device off the secondary, that's what makes this novel. that's why this way is cool and cheaper. You can always regulated down from the 5v rail.