Data for Couchbase (assuming linux) is stored at:
/opt/couchbase/var/lib/couchbase/data
When you say 'too many recorders into the bucket' what do you mean? Just because you have data that is larger than your ram amount doesn't mean the node should go down, it just means that not all documents will be available in ram and the node will sometimes have to fetch from disk (slower). Unless you've exceeded the cluster capability by a huge margin I don't think this is the cause.
I would investigate before wiping the data on the node + backing the data up.
You can also look at the logs which are located at:
/opt/couchbase/var/lib/couchbase/logs
How much ram does each of your servers have and how much is assigned to Couchbase?