Or it means that the formalism of a utility function does not fully describe your preferences.

That is, asking "how much do you care about X", and getting some real number as the answer to that question for any value of X, will not describe the preferences and choices of the agent in question. (This is one way to interpret my previously offered "chickens vs. grandmother" conundrum.)

A more apt formalism might be some sort of multi-tier system, perhaps. I haven't settled on an answer, myself.