While computer models show that the average neutral mutation dies out according to a certain formula, the usual bell curve applies. About half the cases do better than average, and some cases do two standard deviations better. You're right, pass a certain threshold of success and you don't die out unless perhaps your whole species takes a big hit. Meanwhile, new neutral mutations are always occurring.
So what you see when you sample a sexual species is a swarm of alleles for most genes being combined and recombined in unique individuals. Because of all the variation to select from, it's a situation ripe for evolution.
The blind drunkard in a corridor ;)
Will he return to the wall he just lost contact with or will he stagger around until he hits the opposite one?