I agree, there's a significant measurement uncertainty with the manual calculations, and that may be all that I'm seeing. If it averages out over more tanks, then there's no problem.

But a 2 MPG (5%) difference seems a bit too much to explain that way, and it strikes me as odd that it is consistently off by the same amount and in the same direction (so far - 2 tanks is not enough for reliable stats). To me, that suggests that the gauge is not correctly zeroed.