The lack of this feature just caused us to lose huge amounts of data in our production HDFS. Because of a bug in a script, it did a "hadoop fs -rmr /user/hive/warehouse/$variable" when $variable was blank. Its goal, really, was to rmdir /user/hive/warehouse/$variable, and those semantics would have worked under normal conditions, and died under pathological conditions.
The rmr semantics mean the script thinks it succeeded at its task when in fact, it wiped out our cluster. Now I need to spend the rest of my evening restoring from backups (which were, thankfully, not in HDFS).
Could we implement rmdir? It removes the directory iff the directory is empty. This is a far less dangerous operation in the face of scripts with badly-initialised variables.