Especially, dynamic partitioning can create very small files on the file system. It would be useful (and it would reduce the load on the name node) if the files of a partition can be combined into a single larger file.
In particular, the ORC file format supports a very efficient "concatenate" operation, that does not require running a (MR) job. This is surfaced in Hive as the ALTER TABLE ... PARTITION ... CONCATENATE.
One tricky question is whether the concatenation will be atomic, especially, what does this mean for job that are running concurrently with the concatenate operation.