Binormal isn’t the maximum packing efficiency. You can get yet more efficient by adding another smaller size of sphere, to fit in the gaps still left in between size 1 and size 2. And then yet more by adding yet another, and so on. As the number of sphere sizes available increases, the packing efficiency approaches 1.
I’d like to offer my suspicion that that outcome is a function of how they are abraded over time in water. Piled atop each other in a bed, I would think water flow is not uniformly distributed across each stone’s entire surface. The surfaces of the stones affect the water flow around them, which I think would be less of a problem inside a tumbler.
Just guessing; I’m a programmer, not an engineer.