This is how it is - I dropped the whole keyspace but the space was not restored due to *gc_grace_seconds* thing which is standard - 10 days. How to clear the space immediately? Tried nodetool flush, compact, restarting the node - didn't help. It's a single-node cluster, no worries about synchronization.
A guy from the cassandra IRC helped me - I just need to remove all the old snapshots, like that:
nodetool -h localhost -p 7199 clearsnapshot