Normally it is enough to run each check once, repair the problem if it gets detected and remember, that the problem has been checked (that's how it works behind the scenes).
Problems arise if such a "repaired database" is loaded again into an older, still buggy ARB-installation, which then again may break already fixed data. Afterwards your newer ARB-installation will assume things are already fixed and fine, while the opposite is true.
Such a scenario might happen e.g. if you give a database to a colleague who does some modifications and then brings the database back to you.