Understanding the factors associated with variation in the implementation of educational programs has emerged as a key research focus, in an effort to identify how promising programs can be scaled up across their targeted contexts. The current study explored demographic-related variation in program dosage using a novel source of implementation data from the treatment group in a randomized controlled trial (RCT) of Word Generation, a curriculum designed to support deep reading comprehension and academic language. Both student- and teacher/classroom-level predictors were associated with variation in the proportion of curricular activities students completed. We also evaluated how variation in treatment-group program dosage distinctively predicted growth on the RCT's key outcome measures of taught vocabulary words, academic language, and deep reading comprehension. We conclude that design and evaluation of future RCTs would benefit from procedures to predict implementation variation and its consequences prior to randomization.