In a clustered GeoServer setup, the same NetCDF file will be accessed by different instances.
By default the NetCDF store will use an h2 database to keep the NetCDF indexes and by default that database will not be in "AUTO_SERVER" mode.
So the first instance accessing the store will get a lock and will be able to access the data, while the others will fail.
This locking issue can be solved by instructing the store to store the indexes in an external database.
Creating a "mddatastore.properties" file alongside the NetCDF file, will point the store to the database to be used.
The documentation does not mention this and whould be updated.
I don't know if the "AUTO_SERVER" parameter can be specified at all so I did not test it in a clustered environment, but I suspect the instances will have to be able to talk to each other for that to work (the first instance to open the database will act as a server for the others)