- Context. A long-standing problem of strong-lensing by galaxy clusters is the observed high rate of giant gravitational arcs that are not predicted in the framework of the “standard” cosmological model. This is known as the “arc statistics problem”. Recently, several other inconsistencies between the theoretical expectations and observations have been claimed regarding the large size of the Einstein rings and the high concentrations of few clusters with strong-lensing features. All these problems consistently indicate that observed galaxy clusters may be stronger gravitational lenses than expected. Aims. We aim at better understanding these problems by comparing the lensing properties of well defined cluster samples with those of a large set of numerically simulated objects. Methods. We use clusters extracted from the MareNostrum Universe to build up mock catalogs of galaxy clusters selected through their X-ray flux. We use these objects to estimate the probability distributions of lensing cross sections, Einstein rings, and concentrations for a sample of 12 MACS clusters at z > 0.5 from the literature. Results. We find that three clusters in the MACS sample have lensing cross sections and Einstein ring sizes larger than any simulated cluster in the MareNostrum Universe. We use the lensing cross sections of simulated and real clusters to estimate the number of giant arcs that should arise from lensed sources at z = 2. We find that simulated clusters produce ∼50% less arcs than observed clusters do. The medians of the distributions of the Einstein ring sizes differ by ∼25% between simulations and observations. We estimate that the concentrations of the individual MACS clusters inferred from the lensing analysis should be up to a factor of ∼2 larger than expected from the ΛCDM model because of cluster triaxiality and orientation biases that affect the lenses with the largest cross sections. In particular, we predict that for ∼20% of the clusters in the MACS sample the lensing-derived concentrations should be higher than expected by more than ∼40%. Conclusions. The arc statistics, the Einstein ring, and the concentration problems in strong lensing clusters are mitigated but not solved on the basis of our analysis. Nevertheless, owing to the lack of redshifts for most of the multiple image systems used for modeling the MACS clusters, the results of this work will need to be verified with additional data. The upcoming CLASH program will provide an ideal sample for extending our comparison.