Clearing the cache alleviates the symptoms for now.
http://www.elastic.co/guide/en/elasticsearch/reference/current/indices-clearcache.html
Clear a single index
curl -XPOST 'http://localhost:9200/twitter/_cache/clear'
Clear multiple indicies
curl -XPOST 'http://localhost:9200/kimchy,elasticsearch/_cache/clear'
curl -XPOST 'http://localhost:9200/_cache/clear'
Or as suggested by a user in IRC. This one seems to work the best.
curl -XPOST 'http://localhost:9200/_cache/clear' -d '{ "fielddata": "true" }'
Update: these errors went away as soon as the cluster was moved to a faster hypervisor