This logic was present in the old model and keeps e.g. Baloo file search results from getting higher priority than likely more useful application results with the given name. It only affects the order of categories, e.g. scoring "Applications" higher than "Documents" if such a match exists. The order of matches within a category is untouched and remains determined by the runner based on type and relevance.
Since the "relevance" of a QueryMatch is global and determined freely by the runners which don't know what other results might already be present, we have to resort to some fuzzy logic like this.

Searched for "quassel", got quassel application higher rated than my random sprint note file found by Baloo to contain the word quassel.