The ideal answer for this question is: A corpus stays relevant forever once it is introduced into scientific study. There are now repositories for language data that are aiming to keep corpora available for a long time.
For a more practical point of view, large research funding organisations (like DFG in Germany) demand keeping research data available for at least ten years.
Of course, this answer does not necessarily apply to your research project, a corpus from 2004 may be unsuitable for your specific research question (How does the Dutch youth talk in the 2020ies?), or the given corpus may be sampled according to criteria that don't fit to your research question (typically: some metadata you are interested in are just missing, or some interesting registers aren't represented at all). In addition, for an old and established corpus, your research question is maybe already studied with some published results, and you really need a new corpus to see whether the results hold in the new corpus as well.