The bulk write splitting logic for MongoCollection#bulkWrite, because it uses OP_MSG document sequences, is intended to split bulk writes at 48MB, but it actually is splitting at 16MB. The root cause seems to be a conditional that was added back in 2019 as part of a CSFLE ticket. If that conditional is removed, it works correctly (though a couple of CSFLE tests start to fail).

It does not appear to affect the new client bulk write method, which has test coverage for proper split points.