The goal of this paper is to show that evanescent plane waves are much better at numerically approximating Helmholtz solutions than classical propagative plane waves. By generalizing the Jacobi-Anger identity to complex-valued directions, we first prove that any solution of the Helmholtz equation on a three-dimensional ball can be written as a continuous superposition of evanescent plane waves in a stable way. We then propose a practical numerical recipe to select discrete approximation sets of evanescent plane waves, which exhibits considerable improvements over standard propagative plane wave schemes in numerical experiments. We show that all this is not possible for propagative plane waves: they cannot stably represent general Helmholtz solutions, and any approximation based on discrete sets of propagative plane waves is doomed to have exponentially large coefficients and thus to be numerically unstable. This paper is motivated by applications to Trefftz-type Galerkin schemes and extends the recent results in [Parolin, Huybrechs and Moiola, M2AN, 2023] from two to three space dimensions.