First, the example you give is unusual in that molecular dating analysis should be calibrated with fossils to set minimum ages. If Ecrobia grimmi exists in the fossil record, then no proper molecular analysis would estimate them to be younger than those fossils. There are two main causes of discrepancies between fossil and molecular dating.
For fossils you never find the oldest example of a species. If a species diverged 100 million years ago, you might not find a fossil of that lineage for millions of years, meaning that fossil estimates are almost always underestimates, but they do form a baseline upon which to calibrate molecular data. The problem with molecular data is that there is no "molecular clock". Lineages, genes and individual nucleotides all change at different rates. The best you can do is plug in fossil and/or geological dates into a model that is based on other lineages. Hopefully you have calibrated dates for closely related species, and the same genes. Even then, you do not know that the lineage you are working with will follow the same rates as the model predicts. At a minimum, you need to set a minimum possible age in BEAST to 5-16 mya.