Decombination is a useful operator in settings where one has to forget or retract previously given information. A typical case where this happens is when some inconsistency between information sources is observed, where removing potentially conflicting and unreliable information can restore consistency. Other cases where the removal operator is useful include inferences in graphical models and valuation-based systems, where the decombination operator is routinely used. In each case, repeated decombinations of different pieces of information has to be performed, which is conflicting with the fact that, in belief function theory, such an operator is computationally very costly. In this paper, we show that the decombination operation can actually be performed efficiently, at least as long as the number of focal elements remains limited.