A write is considered majority commited when majority of nodes has confirmed to be at least at its timestamp.
So secondary does not have to process it but it has to be queued in the node for processing aka written to its oplog.
If fhe replication oplog window is small it risks one or more secondary to get out of sync if they cannot keepup within the given oplog window. Once they get out of sync they are not counted for majority commit, therefore the chance of failure of w : majority gets higher…