KFS won't really be interesting until they do a MapReduce-clone. You need something like that in order to be able to do anything with the data stored.

they don't need to. it already integrates with hadoop

http://lucene.apache.org/hadoop/

from the perspective of early startups, is this even on the radar since Amazon's S3 is essentially the same thing except without the fuss of buying and setting up your own hardware?