This error is related to the active space (AS) seeder_list which is part of the metadata inside shared nothing store file, it's already recorded with more cache nodes than the initialized ones. i.e. The seeder_list has recorded 4 cache nodes and when restarting the cluster the initialized cache nodes are just two, this will throw the next error:

Recovery fails with not_enough_hosts error

To recover from the error it is necessary to start with all seeders listed in the seeder_list section of the error message. If starting all the seeders is not an option then start with DATA_LOSS recovery option (this article just explains the seeder_list option). Please refer to documentation below with all details on recovery strategies:

https://docs.tibco.com/pub/businessevents-enterprise/6.3.0/doc/html/Default.htm#Configuration/Shared-Nothing-Persistence-Recovery-Options.htm?Highlight=recovery

If you do not want to start with the recovery strategy "DATA_LOSS" then only option is to start with quorum matching the seeder_list entries and set min-seeders property to lower value.  In example scenario with seeder_list 4, you need to start all 4 cache nodes to satisfy the seeder_list or start with recovery policy DATA_LOSS. You can also set quorum to 4 and min_seeders to 2 which essentially means you need 4 cache on startup but post startup you can live with 2 cache nodes. See the steps below:

1.- Come up with 4 cache nodes successfully
2.- Gracefully shutdown 2 cache nodes which you do not need
3.- Let the redistribution happen on the 2 existing cache nodes i.e. seeder_list now contains only 2 cache nodes
4.- Kill the remaining two cache nodes after gracefully stopping inference agents

This way next time cluster can come up with quorum 2 and the two cache nodes that were running last should start successfully.

Note: 

Recovery of system space, such as MasterId or user space, fails with "SYS_ERROR Not_Enough_Hosts" when restarting Cache cluster without any changes: