Good points. Agree that "always go for a big push instead of incrementalism" is waaayyyy too simple and sweeping a lesson to draw from PEPFAR. Also, three cheers for not lying. I think the World Bank was right not to suppress its data on the low cost-effectiveness of ARV drugs circa the mid-2000s. But in retrospect, I think people drew bad policy conclusions from that data.

My piece above is largely a plea for a little bit of intellectual humility and introspection on the part of the cost-effectiveness crowd (of which I'm often an active participant). If we find our analytical framework leads us to oppose the core of PEPFAR (i.e., the distribution of free ARVs in poor countries), then we ought -- in the face of its historic achievements -- to pause and reflect a bit, rather than shrugging and insisting "yeah, but I could've done better." 

Your first paragraph gets at one reason why I think that kind of confidence isn't credible. Successful trials don't automatically translate into successful national or multi-national, multi-billion dollar programs. So we should be more skeptical that something that scored higher on cost-effectiveness terms ex ante would've converted into a massive global policy win. I genuinely don't think cost-effectiveness analysis, at least as practiced in global development/global health, really takes those questions of 'scalability' and implementation feasibility seriously enough. As a result, we confuse stuff that really could be done a million times over if money was available with stuff that looks cheap but faces much deeper political and organizational hurdles.

When we find something that works and really can scale with an influx of cash, we should run with it. With the benefit of hindsight, PEPFAR seems to fit that description. And, coincidentally, setting aside broader philosophical debates to stick with this concrete case, PEPFAR happens to be up for reauthorization in congress this year...

At the risk of over-simplification: I think you should at least discount the expected value of all those alternative health mega-projects by (a) the probability that congress would've funded them, and (b) the probability they would've worked as intended at scale. If the first probability was, say, 10%, and the latter maybe 25%, numbers get pretty small pretty fast. (Of course, PEPFAR only scores 100% and 100% in retrospect. But it seems it would've still scored quite high circa late 2002.)

Longer version...

1. Fully agree that $100b investment in malaria eradication might have changed a whole bunch of parameters too. This is a good point. But it also reinforces the point that static, comparative cost-effectiveness analysis is a bad framework for making these choices. What frustrates me is that even after 20 years, we lack a good analytical framework for thinking about why PEPFAR worked so well.
2. I didn't harp on this point in the original post, but some of the "better" alternative uses of PEFAR money that were put forward both then and now -- e.g., HIV prevention rather than treatment --  lack the same kind of proven effectiveness at scale that we've seen for ARVs. Development is littered with 'proven' interventions that fall apart at scale. Did the White House get lucky in picking something with sufficiently simple logistics that it could work across dozens of countries and millions of patients? Or did Paul Farmer, Mark Dybul, etc know something we don't? Not sure.
3. It's always difficult to know whether to build politics into the model when doing economic calculations. I'm sure there are instances where saying "I know you (policymakers) don't want to do it, but policy A would be way better than B."  PEPFAR just illustrates how much money we potentially leave on the table by respecting that partition between politics and technical advice. If the doctors summoned to the White House had lectured the policy folks on how a big push on HIV treatment was irresponsible, would they realistically have gotten something better? Who knows, but there's little in the historical record of PEPFAR's genesis to suggest that was a real possibility.