We present an algorithm for efficient 3D inversion of marine controlled-source electromagnetic data. The efficiency is achieved by exploiting the redundancy in data. The data redundancy is reduced by compressing the data through stacking of the response of transmitters which are in close proximity. This stacking is equivalent to synthesizing the data as if the multiple transmitters are simultaneously active. The redundancy in data, arising due to close transmitter spacing, has been studied through singular value analysis of the Jacobian formed in 1D inversion. This study reveals that the transmitter spacing of 100 m, typically used in marine data acquisition, does result in redundancy in the data. In the proposed algorithm, the data is compressed through stacking which leads to both computational advantage and reduction in noise. The performance of the algorithm for noisy data is demonstrated through the studies on two types of noise viz. uncorrelated additive noise and correlated non-additive noise. It is observed that in case of uncorrelated additive noise, up to a moderately-high (10%) noise level the algorithm addresses the noise as effectively as the traditional full data inversion. However, when the noise level in the data is high (20%), the algorithm outperforms the traditional full data inversion in terms of data misfit. Similar results are obtained in case of correlated non-additive noise and the algorithm performs better if the level of noise is high. The inversion results of a real field data set are also presented to demonstrate the robustness of the algorithm. The significant computational advantage in all cases presented makes this algorithm a better choice.