The deep water data is a challenge for predictive methods like deconvolution or the upward continuation used here, because of its multiple periodicities. I find that both multiple trains can be removed together, but there are a large number of adjustable filter coefficients which raises the risk of removing primary energy as well as multiples. It should be interesting to see how moveout-based algorithms compare, because they do not have the problem of choosing a multiple period, though because some multiple paths spend more time below the seafloor, the normal problem of resolving differences in moveout may be more difficult.
The shallow water data poses a stubborn problem, because the multiples are so bright and closely spaced. Some energy does make it out of the suppression algorithm, but it is not coherent.