The COMMIT NFSv4 (and NFSv4) request asked to sync a range of a file. However the Linux VFS only allows a whole file to by synced
Some filesystems, certainly XFS, can implement a subrange sync easily. It should be more efficient, but until that can be demonstrated, it might be hard to push the change into the VFS layer. Then again, it might just need someone to be persistant.
A suitable benchmark would probably involve writing a very large file while there was lots of other activity on the filesystem. With a more efficient COMMIT, the COMMIT request should never block.