Hashes vs key scan in REDIS

My team is using Redis for caching and I want to cache the following structure:

inbox --(1 to many)-> folders --(1 to many)-> messages

It's just a simple tree with no nesting, and each inbox/folder/message is just a single int/long. I more often need to get all messages in a single folder, but sometimes also need to get all messages in a whole inbox (all the folders). For the Redis data structure, I'm considering the following 2: Storing messages in a list & using SCAN

key = "inbox_id:folder_id"
value = [message_ids]

fetch_folder(): GET inbox_id:folder_id
fetch_inbox(): SCAN inbox_id:* | GET resulting_ids
// I'm not sure how to do fetch_inbox in just Redis yet, so pipe for now

Having a hashmap per inbox

inbox_id = hashmap { key = folder_id, value = "message_ids_as_string" }

fetch_folder(): HMGET inbox_id folder_id
fetch_inbox(): HVALS inbox_id

Which storage & access pattern would be more performant, assuming 10-100 folders per inbox & 100-10_000 messages per folder?