Suppose you have a system of interacting variables with some "best" (Pareto-optimal) assignment of values to variables, say {A,B,C} and {A=5,B=10,C=20}. Suppose also that some external constraint prevents the realization of this best solution, e.g. A can't be lower than 10. The second best solution is not necessarily obtained by fixing the other values, e.g. {A=10,B=10,C=20}.
The implication is that in a complex system with multiple dysfunctions, fixing one of the dysfunctions without touching the others may actually make the overall function of the system worse.