Do you mean to say that the expected inclusive fitness of a randomly selected creature from the population goes up with time? Well, if we sum that up over the whole population, we obtain the total number of offspring - right? And dividing that by the current population, we see that the expected inclusive fitness of a randomly selected creature is simply the population's growth rate. The problem is that evolution does not always lead to >1 population growth rate. Eliezer gave a nice example of that: "It's quite possible to have a new wolf that expends 10% more energy per day to be 20% better at hunting, and in this case the sustainable wolf population will decrease as new wolves replace old."