Using three models of upper mantle hydration state, we quantify the subsolidus process of mass redistribution within and out of slabs. We use a numerical model where all major elements in the rock are soluble and mobile. Our transport model achieves high accuracy in predicting the energetics and fluid/rock mass balance of subduction zone processes. We find that most carbonate rich sediments (e.g. eastern Pacific) get depleted in K while they accumulate Al and Si. Patterns of mass and concentration changes are not always correlated, reflecting the antagonistic role of metasomatism versus residual enrichment due to volatile loss. We find that carbon released to subarc depth is 25.7 – 31.4 Mt C/yr, which corresponds to 51.0 to 61.9 % of trench input, confirming previous estimates. However, this decarbonation efficiency is highly heterogeneous. The host slabs of eastern pacific recycle most carbon back to the mantle wedge, while the cold slabs of western pacific inject carbon. Slabs decarbonation efficiencies show different sensitivity to hydration state of the mantle wedge. Transport (water)-limited slabs, which are sensitive to the AUM hydration state, are the most numerous and tend to be the largest C emitters. There are also supply-limited slabs, i.e. not sensitive to the AUM hydration state, e.g. hot (Mexico) and cold (Honshu) slabs. We first show that slabs subducting limestones are sensitive to infiltration of non-volatile elements from underlying igneous sections to drive metasomatic-decarbonation process by promoting garnet growth. These insights inform specific regional targets for studying the cycle of rock-forming elements.