I do not have inside knowledge of the monday.com GraphQL resolver, I’m only speaking based on observation and reports publicly. All assertions here are my own speculation though presented as absolutes so I don’t have to say appears 50 times.
It appears the problem is death by a thousand fetches in the resolver. The current resolver is iterating items, and then the items columns, and fetching for each column. Thats the only way that explains the growth by mirror count * item count.
I suspect a tenacious programmer could devise a better resolver that iterates many items and columns, and builds a large list to fetch, and then caches the results because it is not uncommon for a board of many items to link to the same small set of items (such as orders to customers, clients to sales reps, or employers to states).
An ambitious one could probably figure out how to asynchronously do this in some fashion.
Knowing what you can get from the API about mirror columns, everything is there to implement a more efficient proxy resolver locally. This seems like, if monday can’t promise a solution in a reasonable timeframe, a potential albeit costly and painful workaround.