Yes, I believe it's a bug (see my question on the user lists: https://mail-archives.apache.org/mod_mbox/giraph-user/201403.mbox/%3COF416E2CF4.1613A751-ON86257C9F.00498FA5-86257C9F.0049D454@us.ibm.com%3E). The way I'm getting around it is by sending a kill to all the workers after a job is done:
ssh worker-name "kill -9 \$(ps aux | grep \"[j]obcache/job_[0-9]\{12\}_[0-9]\{4\}/\" | awk '{print \$2}')"