2 immediate thoughts -
Firstly, in terms of human welfare per economic value, the graduation approach is probably more effient. The received by graduates is received by people who were previously in poverty (and people close to them, particularly their children). I expect that Growth in general, like that experienced by China in the Deng Xiaoping period, is less efficiently distributed than the graduation approach. But I expect the efficiency factor is less than 10. So Hallestead and Hillebrandts position stands that critique.
Secondly, H&H strawman the randomista position. Dufflo and Bannergjee argue in Poor Economics that the gains from effective charities are large relative to regular charities. But more importantly, the randomista development can shift the policies of development countries on important issues below macro-economics. The growth potential to changing textbook purchasing in India through RD could compete with development economics. If a RD study leads to new textbooks, tens of millions of children would read them per year. That lever is comparable to the lever described by H&H (one study to millions of children receiving an education-year-equivalent). For more of this perspective, check out the 80k interview with Rachel Glennerster - http://documents.worldbank.org/curated/en/243261538075151093/pdf/WPS8591.pdf
The second point defends IPA, JPAL, One Acre fund from H&H's critique because each produce research outputs that may change developing country policies (da big lever). For the malaria and give directly approach H&H's critique stands.