A distributed file system intelligently allocates portions of a file system from a server to one or more clients on a network. File system data used during power-on of a client may be stored on the client. A retrieval of the file system may subsequently be made to the extent the client has capacity to store the file system. A multicast operation, performed as a background operation on the client, may be used to retrieve the file system. For portions of the file system not allocated to the client, the client may still access the file system from the server. Network bandwidth may thus be reduced and client access to the file system may generally be more efficient.