You can try out MapR's Hadoop distribution, which uses its own filesystem rather than HDFS. The MapR filesystem is not append-only, handles small files well, is easily accessible over NFS, etc.

(*Note: I'm a MapR employee, so obviously I'll be biased towards thinking our stuff is great)