The cumulative reaction probability [CRP or N(E)] for the four-atom reaction, H-2+OH-->H +H2O is calculated using one of the formulations of Miller, Schwartz, and Tromp [J. Chem. Phys. 79, 4889 (1983)] and the transition state wave packet (TSWP) approach of Zhang and Light [J. Chem. Phys. 104, 6184 (1996)]. It is shown that locating the dividing surface of the flux operator in the transition state region significantly reduces the number of wave packets which must be followed in order to converge the CRP as compared to the use of initial state selected wave packets (ISSWP). In addition we examine the use of transition state normal coordinates (versus Jacobi coordinates) and show that the use of transition state wave packets defined in normal coordinates yields more rapid convergence of the CRP and individual contributions of the TSWP to the CRP can closely approximate the probabilities of reaction for each transition state as a function of energy. Problems with large amplitude motions using the normal coordinates of the loose non-linear transition state are shown to be absent if normal coordinates of a linear transition state are used. Applications to the 3-D H + H-2 (J = 0) reaction and to the 6D H-2 + OH (J = 0) reaction demonstrate that both N(E) and the initial state reaction probabilities at many energies can be evaluated accurately and efficiently by propagation of each TSWP only once.