One hundredth of an ounce in 8 ounces is one part in 800 or 0.125% - say ten bit accuracy.
This sort of stuff is never particularly cheap.
Farnell lists a 3kgm (6.6 pound) load cell for 194.94 euro (about $260) which is probably accurate enough to do your job.
What you probably want is an open-pan laboratory scale with a 0 to
250gram range
formatting link
but they cost even more, and are more accurate than you need.
To get the precision you want, you'd probably need to balance the weight of the object by the repulsion between two current-carrying coils, where the moving coil is tightly contrained to move only up and down in relation to the fixed coil so that it neither tilts nor slides sideways - I'd think in terms of berylium copper leaf springs, slightly offset, so that vertical movement would be accomated by a minimal rotation of the moving coil.
If you can keep the area free of ferromagnetic materials (their permeability is temperature dependent) and monitor the current through the coils (and not the voltage drop across the coils, because the resistance of the coils is also temperature dependent) you could get your accuracy easily enough. I'd use a capacitance sensor to monitor the height of the moving coil - traditionally one used to use an optical lever, but capacitance is much nicer.
------------- Bill Sloman, Nijmegen