I wouldn't go with a dilution, I'd simply make a thinner sample tube, maybe like those hydrometers for measuring antifreeze.
You could calibrate the tube for each of the balls (sink/float) to get big divisions, and then interpolate values between depending on how much the partuicular ball...