I don't know what happens at the MongoDB layer (for MongoDB questions, including questions about WiredTiger inside of the MongoDB engine, please use the
MongoDB Community forums.)
That said, WiredTiger itself always prewarms the cache by loading the first internal page layers of each Btree on open (I think the argument there is (1) there aren't a lot of internal pages at high-levels of the tree, (2) they're almost certainly going to be useful, where as leaf pages might or might not be useful).
I haven't thought about this hard, but I don't think WiredTiger would actually need to "dump the cache"; the on-disk code we use (we call it the "block manager"), knows the pages that were recently written by reviewing the last checkpoint information for the file. While that certainly wouldn't be useful for all workloads, it wouldn't be difficult in WiredTiger to figure out the pages that were recently written and read them back into cache on startup.