Note that GiveWell carefully tracks room for more funding in its charities. They channeled funding to VillageReach until it had a few years worth of funding at the margin, and then moved on in their recommendation. But they keep track, and when VillageReach again shows that it can use money effectively it will be recommended easily. The problem with Haiti was that there were a lot of donors giving because of the TV images and charity fundraising even though it was clear that there was a surfeit of funding.

If those donors had been more sensible they could have used the recommendations of a service like GiveWell to identify the top organizations. Likewise for scientific research one can back a lab or fund that can allocate resources among many different experiments (or unrestricted grants to Doctors Without Borders which it can allocate to locations of greatest need). This doesn't seem to be too much of a problem in practice, as well as theory.