Thanks for reporting this issue. We're taking a look at it and considering improvements.
For now, all I can recommend is to retry the operation if it fails, with a slight delay since the problem may simply be that the cluster is busy.
You might also want to consider smaller deletes, using narrower queries that select fewer documents/rows.
I'll check into whether you can reasonably tune the timeout to handle cluster latency issues.
EDIT: You can raise the timeout with write_request_timeout_in_ms, in cassandra.yaml