An anonymous commenter pointed out this quote from Duflo, Dupas, & Kremer (2011) - in point 1 above: "While tracking increases test scores for students at all levels of the pretest distribution assigned to be taught by contract teachers… initially low-scoring students did not benefit from tracking if assigned to a civil-service teacher”. So ability tracking all by itself didn't have the positive benefit; it needed to be combined with the accountability intervention (point 3 above). I also discovered another paper after writing this, by Duflo et al. (2015) [http://www.3ieimpact.org/media/filer_public/2015/02/24/ie_22_evaluation_of_cce_and_lep_in_haryana.pdf] From a cursory glance, it seems that once again there are gains to a form of ability tracking: "students are given a brief assessment of basic Hindi skills at the start of the academic year, and a portion of the school day is set aside to group and teach students according to ability level, regardless of age or grade." Once again, this isn't plain ability tracking. It's a very particular format. But it IS a form of ability tracking, giving the teacher a chance to target teaching to the students' ability levels.