This and related questions were thoroughly discussed about 12 years ago on alt.coffee by David Ross, who is a statistician.

Beans of different size and density might segregate; for instance, a blend with denser light roasted beans and less dense dark roast ones might end up with the denser smaller light roast beans at the bottom. However, I don't recall hearing any actual cases; although it sounds like you may be one.

The more prevalent problem is simple random variation in the proportions of beans shot to shot, even assuming the beans are well mixed in the hopper. If you are emaking double shots, it's not a poblem if the blend has two or three ingredients at roughly 50/50 or 33/33/33 proportions. But when you get blends with lots of beans, or with beans that are less than about 15%, all bets are off. A double shot contains about 90 beans, and the bean variance can be calculated using the binomial distribution; a 10% bean can end up anywhere form 0% to 20% shot to shot.

Italian blends which have 10% to 20% robusta are crap shoot if you are grinding single portions. When a cafe uses a filled doser, the problem goes away, since the doser remixes the ground coffee. Portion grinding is a good reason for third wave blenders to stick to two or three roughly even ingredients in their blends.