There's an important caveat with the mediacounts data. Originally we had a rough design of the media request metrics (wikitech) in which, much like the page view metrics, they would have project as a dimension. However, the mediacounts dataset has no idea of which project is using the requested file. A Commons image could be requested from Wikipedia or Wikispecies, but it could also not be used at all from a wiki, and instead be hotlinked from a non-wiki page.
So, as it is, the mediacounts data can't be directly split by project. However, the URIs in the dataset contain data about which wiki was the file uploaded for. According to the definition of the mediacounts dataset, these are all the possible ways that an upload.wikimedia.org URI can start with:
/math/* /score/* /wikibooks/{language}/* /wikinews/{language}/* /wikimedia/{language}/* /wikipedia/{language}/* /wikiquote/{language}/* /wikisource/{language}/* /wikiversity/{language}/* /wikivoyage/{language}/* /wiktionary/{language}/* /favicon.ico/*
Where {language} can be a valid ISO language code as used in wiki URLs (en, be, ca) or commons.
Examples:
/wikipedia/bar/timeline/79799ec4287e24767726404d81fc8897.png
/wikipedia/commons/0/00/Ancient_Egypt_map-el.png
/score/l/4/l42r5igkbg59dqectsgpc0807oh7of3/l42r5igk.png
/math/e/9/4/e94049b807364f202efe747fd69f247e.png
So we could expand again the GetMediaFilePropertiesUDF to add info about the project/namespace.