Consolidating the S&H into the item cost makes sense for a subset of situations sure, but there's a huge set where it's exploitive.
Amazon orders for example often can amortize S&H costs by shipping a single package containing multiple items sharing a common origin. I'd argue that's true for a huge portion of Amazon orders. It's simply not a good generalization.
Having a fixed per-item shipping costs lowers overhead drastically, especially for smaller sellers. Large sellers obviously have different customers and I doubt they get away with just multiplying PER_ITEM_SHIPPING_COST X NUMBER_OF_ITEMS (I didn't think that was the focus on the OP though).
I also disagree that it does not generally apply. Given two items, A & B, you might have the following total:
A_shipping + A_price = A_total
B_shipping + B_price = B_total
What you're saying is that: AB_shipping + A_price + B_price < A_total + B_total
But what I'm saying is that: A_shipping < AB_shipping
B_shipping < AB_shipping
In other words, no matter what, shipping more items does cost more than shipping less items. So it makes sense to charge more for it. How much more is a matter of economics.