There has been one evaluation of Rosetta Stone that I know of. It isn't published, to my knowledge, but you can find it on the internet: "Measuring the effectiveness of Rosetta Stone" by Roumen Vesslinov, Queens College.
The claim is that 55 hours of Rosetta Stone, doing nothing else, is worth a semester of college Spanish.
The mean posttest score was 237.7, a gain of 185.1 points on the WebCAPE test. Vesslinov says that the average gain was 4.3 points per hour. That's based on the 237.7 final mean score. Based on gains, the average gain was 3.37 points per hour. This suggests that a genuine beginner would score about 185 points at the end of 55 hours.
According to Vesslinov, a college Spanish program that lasts four semesters requires at least 270 on the WebCAPE to be placed in the second semester (apparently four unit classes). One that lasts six semesters (apparently three unit classes) requires 204 to be placed in the second semester. The 237.7. average final score falls between these two, and considering the pretest score, the average final score of 185 falls short of both.
On the ACTFL oral interview, about 59% improved one level, on a 1-5 scale. But about one-third did not improve at all.
Does RS work? I we accept the 3.37 points per hour figure, it would take 80 hours to reach the second semester threshold.
No comparison group was used. It would be important to have a comparison group with the same background, especially level of education. And of course it would be important not to include false beginners.
There was also no a clear description of the RS program. I have heard their ads, however, about how we shouldn't "learn" a language but we should "absorb" it. I think I've seen something like this before ...
Here is the Rosetta Stone spin on this study: http://rosetta-stone.toptenreviews.com/hard-science-on-rosetta-stone-effectiveness.html. The article claims that 55 hours of RS is as good as 45 hours of class plus 39 hours of homework (84 hours). But remember that the study used a highly educated group, and had false beginners. My estimate is 80 hours, not 55.