To demonstrate this more fully, consider the case of having the wheel flanges on the outside, with the conicity of the wheels pointing the other way. Gravity would still tend to centre this arrangement, but I'm told that if you build such a system in practice, then it won't run nearly as smoothly.

(PhD was 'Residual stress in rails', for what that's worth. Judging from the profiles of the rails I saw, direct contact with the wheel flange plays a substantial role in keeping the train in place on curved track. But on roughly straight track, I'm satisfied that the argument about conicity applies).