These studies consistently find substantial variation in teacher effectiveness. For example, the findings of Rockoff (2004) and Rivkin, Hanushek and Kain (2005) both suggest a one standard deviation increase in teacher quality improves student math scores at least 0.1- 0.15 standard deviations. Aaronson, Barrow and Sander (2007) find similar results using high school data.
Indeed, the inherent optimism of this literature is captured by an oft-cited statistic that matching a student with a stream of good teachers (one standard deviation above the average teacher) for 5 years in a row would be enough to complete eliminate the achievement gap between poor and non-poor students (Rivkin, Hanushek and Kain 2005).
I don't understand why people get all giddy over a 0.1 - 0.15 standard deviation, i.e., educationally insignificant, gains. Gains like this just don't show up in the real world. This isn't even a real scientific experiment with a control group taught by random teachers and an experimental group taught by super teachers. This study is just the playing of statistical games and drawing conclusions from the tiny correlations.
I suppose they'll have to go to Lake Woebegone to find all these above average teachers -- not that we know what to look for anyway.
In any event, this study seems to disprove the notion that years of successive great teachers aren't cumulative since only about 20% of the gains are maintained.
The entire notion of cherry picking great teachers was silly from the get go.