Assuming Rhino gives you the correct volume, the weight that any software gives you is correct if it uses the correct density (which for jewellery is historically established). All those processes alter the actual volume or surface and there is no easy way to measure the error, because for some of those processes it is related to area and some to volume and some to edges… and they can’t generalise it either because your manufacturing processes are not those of another user…So I think that yes, it is a problem that everyone has, but in some pieces it manifests a lot and in others a little.
I think the best solution is to separate your production into one, two or three levels of complexity (where a sphere is 0 complexity and an extremely sharp shape is very complex), and measure the error of each piece, at least 20 or 30 samples per group. Then for next pieces you just categorise it into the proper complexity-shape group and use that estimated error (an average) for that group. The more samples you use and the more complexity groups you have, the more accurate you can be.