By default, the replication level for job jars, libjars and the distributed cache in general is mapred.submit.replication=10. This yields under-replication warnings for these files on small clusters (less than 10 data nodes) when using fsck ("hadoop fsck") on their HDFS.
Example on an 8-node cluster:
/tmp/hadoop/mapred/system/job_201105191458_1857/job.jar: Under replicated blk_-6996370258385460742_366223. Target Replicas is 10 but found 8 replica(s).