We present a new approach to factorize and resum the post-Newtonian (PN) waveform for generic equatorial motion to be used within effective-one-body (EOB-)based waveform models. The new multipolar waveform factorization improves previous prescriptions in that (i) the generic Newtonian contribution is factored out from each multipole; (ii) the circular part is factored out and resummed using standard EOB methods; and (iii) the residual, 2PN-accurate, noncircular part, and in particular the tail contribution, is additionally resummed using Pad\'e approximants. The resulting waveform is validated in the extreme-mass-ratio limit by comparisons with nine (mostly nonspinning) numerical waveforms either from eccentric inspirals, with eccentricities up to $e=0.9$, or dynamical captures. The resummation of the noncircular tail contribution is found essential to obtain excellent ($\ensuremath{\lesssim}0.05\text{ }\text{ }\mathrm{rad}$ at periastron for $e=0.9$) analytical/numerical agreement and to considerably improve the prescription with just the Newtonian prefactor. In the comparable mass case, the new 2PN waveform shows only a marginal improvement over the previous Newtonian factorization, though yielding maximal unfaithfulness $\ensuremath{\simeq}{10}^{\ensuremath{-}3}$ with the 28 publicly available numerical relativity simulations with eccentricity up to $\ensuremath{\sim}0.3$ (except for a single outlier that grazes ${10}^{\ensuremath{-}2}$). We finally use test-particle data to validate the waveform factorization proposed by Khalil et al. [Phys. Rev. 104, 024046 (2021)] and conclude that its amplitude can be considered reliable (though less accurate, $\ensuremath{\sim}6%$ fractional difference versus 1.5% of our method) only up to eccentricities $\ensuremath{\sim}0.3$.