From a casual perspective, "order of growth" can be thought of as the shape of the worst-case growth curve, regardless of the slope of that shape. Even it a line indicating linear (i.e., O(N)) growth has a high slope, eventually--when the data set is big enough--it will be overtaken by an O(N²) curve. That's one of the pitfalls. People get seduced by the behavior of their code on small data sets, then get blindsided when they try a much larger data set and performance sucks.