I'd think the cutoff would be subtle. I first discovered this phenomenon while working with small simple gif/png images. But they weren't as small or as simple as the 4 x 4 white we just tested. Dimensions, number of colors, dithering, and complexity of the image itself would all be factors I imagine, though it may be more simplistic than that. I'm thinking it is the interplay among these and possibly other factors that makes the difference. With slightly larger, slightly more complex images, I believe the savings with .gif is greater.

And yes, the 60 or so bytes is no big deal, but as the number of images of this sort involved on a page increases, and if they are slightly larger (engendering greater savings per image) . . . especially for slower connections, it could be important.