The classical Rice-Ramsperger-Kassel-Marcus formula is tested at an accuracy level of a few percent by comparing results of numerical phase space integration with lifetimes deduced from trajectory calculations. The test object is HD2+; the calculation has been done for total energies of 0.5, 1.0, and 1.5 eV above dissociation, and for total angular momenta of 0-60 (h) over bar. Presupposing that the trajectory calculations show the true classical dynamics, we find systematic deviations of up to 40% of the RRKM results. They can be fully explained by the influence of "direct trajectories," a special kind of nonergodic behavior of the system. After correction for this phenomenon, both methods agree to within the accuracy of the calculations, which is about 3%. We also verified that the discrepancy vanishes when the energy approaches the dissociation energy.