I know very little so far about the Percona cluster but what I think is happening is that you are making a change outside of what the cluster knows about. The cluster only knows about what you execute through the MySQL interface. Therefore it finds a data problem in the missing directory and recreates it. Instead of deleting a directory to delete the database, if you were to execute a ‘drop database database1;’ it would synchronize and delete as you expect.
In this case you delete a file from the filesystem perspective, so only when you will write data into that table from another node, this node will be stopped because it won’t be able to replicate. Then a full SST will be indeed needed.
You can force SST by removing grastate.dat file and set global wsrep_cluster_address=“gcomm://IP_OF_ANOTHER_NODE”