To limit global warming to well below 2°C, global CO₂ emissions must reach net zero, implying that carbon dioxide removal (CDR) will be needed to compensate for residual emissions. Additional CDR will potentially be required in overshoot pathways that reduce global temperature after a peak in the second half of this century. Here, we explore the efficiency of two marine carbon dioxide removal (mCDR) methods, Ocean Alkalinity Enhancement (OAE) and Direct Ocean Carbon Removal (DOCR), and the carbon cycle response to their deployment in the coupled Earth system. We model the additional deployment of mCDR during 2040–2099 in an overshoot pathway that originally assumes only terrestrial CDR (SSP5-3.4-OS), using four emission-driven Earth System Models (ESMs). Our coastal OAE deployment is based on a life cycle assessment (LCA) of ocean liming, while DOCR is implemented targeting a similar carbon removal as OAE. We define mCDR efficiency as moles CO₂ taken up by the ocean per mole alkalinity added (moles DIC removed for DOCR). These metrics shows a substantial model spread when averaged globally over the last decade of our simulations: 0.53–0.69 for OAE and 0.66–0.94 for DOCR. These values reflect uncertainties in ocean circulation, mixing, and carbonate system biases in ESMs. Our simulations confirm that a significant part of the additional air-sea CO₂-flux is located outside the deployment regions, but we find a substantial model uncertainty also for this metric. Our models indicate that on average 42–57% of the additional CO₂ uptake for OAE and 49-65% for DOCR happens inside the coastal deployment regions. The reduction of atmospheric CO₂ levels compared to the SSP5-3.4-OS baseline varies across models, with a multi-model mean reduction of 7.6 ± 1.6 ppm for OAE and 6.5 ± 1.6 ppm for DOCR in the last decade of our simulations. This corresponds to a modest cooling of 0.07 ± 0.04 °C (OAE) and 0.04 ± 0.04 °C (DOCR) relative to the baseline, with no significant influence on the timing of overshoot recovery. The temperature as well as the land-carbon responses are found to be small and mostly not statistically robust relative to the (modelled) internal variability of the climate system. Overall, our multi-model assessment suggests that OAE and DOCR can contribute to long-term atmospheric CO₂ reduction, but inter-model differences still need to be better understood and reduced to improve estimates of marine CDR efficiency.