Because testing can show the blood to be clean even for someone who does have HIV, and blood/marrow aren't (I believe / assume) things you can just stick in a fridge for a couple of years and then test again.
So if the transfer can work with a 3 month delay (I'm not sure if it can, just assuming), and the HIV test comes back clean, you're going to have a 97% chance of a true-negative, 3% chance of false-negative.
At that point, comparing it to the odds that the "eliglble" donor is lying about their sexual activities or is simply HIV positive and heterosexual, the ban on letting gays register seems draconian, and a scientific test-based approach seems better all around.
Especially when you consider that if no donor is found, the patient could die anyway. So if the only donor is gay, test them, and then notify the patient of the 3% chance of contracting HIV from a false-negative patient versus the 100% odds of dying from Leukemia, and see what decision they make.
I probably screwed something up, but in any case you are definitely right that the false-negative odds are very low among all potential applicants. Really makes the argument against seem foolish.
1) http://www.cbsnews.com/8301-504763_162-57333212-10391704/cdc...
2) http://www.organtransplants.org/understanding/marrow/
3) http://www.wolframalpha.com/input/?i=%28%281%2F20000%29++*+%...