Not to be confused with stone_type=*. 
To be sure, fragmentation is a major downside to refining material=* inline. It places extra burden on data consumers and query developers to know about many obscure values, not just the most common ones. Besides material=*, a variety of keys have values that can be refined in place, such as denomination=*, cuisine=*, operator:type=*, sport=*, and surface=*.
What we really need is a reliable, interoperable mechanism for data consumers to learn that, for instance, granite (16,843 occurrences), sandstone (10,141) and marble (6,745) are each a kind of stone. It is possible to query Wikidata for this information, although a data consumer would need to avoid interpreting the results too literally, because Wikidata’s ontology isn’t always a perfect match for OSM tag definitions. In theory, a more reliable approach would be to query data items and Wikidata simultaneously using Sophox. Unfortunately, Sophox has a bug that prevents it from indexing data items. Also, many data items on the OSM Wiki lack an appropriate Wikidata concept (P12) statement, especially after the push to remove Wikidata from inboxes somewhat obscured this mechanism from users.
Regardless, the important thing is to tag the detail, not lose that detail just because of a disagreement about where to put it.