[achal]

From what I can tell, the actual issue in this case was that they simply summed the wrong thing, rather than ran into some floating point error.

Also, I haven't gone through the first paper, but that second link just seems to show how they generally conform to IEEE 754, except for three cases (div/0 errors instead of infinity, NaN errors immediately instead of allowing further computations, and [this one is a bit more important] they don't implement denormalized numbers). I'm not sure why that would mean Excel is terrible at floating point math.