We study the control of chaos in an experiment on a parametrically excited pendulum whose excitation mechanism is not perfect. This imperfection leads to a weakly excited degree of freedom with an associated small eigenvalue. Although the state of the pendulum could be characterized well and although the perturbation is weak, we fail to control chaos. From a numerical model we learn that the small eigenvalue cannot be ignored when attempting control. However, the estimate of this eigenvalue from an (experimental) time series is elusive. The reason is that points in an experimental time series are distributed according to the natural measure. It is this extremely uneven distribution of points that thwarts attempts to measure eigenvalues that are very different. Another consequence of the phase-space distribution of points for control is the occurrence of logarithmic-oscillations in the waiting time before control can be attempted. We come to the conclusion that chaos needs to be destroyed before the information needed for its control can be obtained.