$ du -sh /data/local/wdong/data # the directory contains a bunch of 30MB files.
15G /data/local/wdong/data
$ time cp -R /data/local/wdong/data . # copy data via fuse
real 3m9.192s
user 0m0.148s
sys 0m19.581s
$ time hadoop fs -put /data/local/wdong/data test/data1
real 2m56.955s
user 0m16.225s
sys 0m30.286s
So whether via fuse or hadoop commandline, the write throughput of mapr is about 80MB/s, with the hadoop commandline being slightly faster. The overhead of java is actually negative compared to that of fuse. I expect the performance of MapR's native NFS server should beat both.
I'm using a $20 TRENDnet 8-port Gigabit Switch and there is a cluster of 7 MapR servers behind it.