The plateau in the duration distribution of long gamma-ray bursts (LGRBs) provides a direct observational evidence for the Collapsar model. The plateau reflects the fact that the observed duration satisfies: T90 = te-tb where te is the time that the central engine operates and tb is the threshold time, interpreted within the Collapsar model as the time it takes for the relativistic jet to penetrate the stellar envelope. Numerical simulation and macronova observations suggest that compact binary mergers involve mass ejection. If short-gamma ray bursts (sGRBs) arise from such mergers, their jets should cross this surrounding ejecta before producing the prompt emission. Like in LGRBs, this should result in a distinct short plateau in the GRBs' duration distribution.We present a new analysis of the duration distribution for the three GRB satellites: BATSE, Swift and Fermi. We find a clear evidence for a short (~0.4 s) plateau in the duration distribution. This plateau is consistent with the expected jet crossing time, provided that the ejecta is of order of a few percent of solar masses. © 2017 The Authors Published by Oxford University Press on behalf of the Royal Astronomical Society.