Consider a Lightning payment: Alice wants to pay 100 sat to Dave via two intermediary hops: Alice - Bob - Charlie - Dave. Both Bob and Charlie advertise a fee of 2 sat. Alice sends 104 sat to Bob, expecting him to forward 102 sat to Charlie. But Bob only forwards 101 sat. Then Charlie has a choice: either to fail the payment and get nothing, or forward it for just 1 sat. It seems that the economically rational choice is to forward anyway. If this is true, why first hops not always take nearly all fees for themselves?
Of course, due to onion routing, Bob doesn't know whether Charlie is the last hop. If Charlie is the ultimate recipient and he won't get the sum he expected, he won't reveal the preimage, and Bob will get nothing. But can such strategies be profitable on average, over many attempts?