Why set the authorized_keys file to a local pathname on large UNIX environments, especially when NFS shares are used for home directories? Because this can address security problems.

First, you must remember that this special SSH file stores the public key of a remote account, letting the owner to be able to log-in using asymmetric keys along with the corresponding passphrase instead of the more classical challenge with appropriate password mechanism. (This eventually enable for non-interactive login through the use of an SSH agent, latter.)

The default path for the authorized_keys file is in a subdirectory of the home directory. This means that when the home of a UNIX account is hosted on a NFS share, all servers available in the same domain as the NFS resource will have access to the very same authorized_keys file, thus opening a security flaw. This is a security concern since by allowing one account on one server, you open this account to all servers in the same domain.

So, the first benefit to store the authorized_keys file in a local name space on each server is to authorize one--and only one--access to a given machine. The direct drawback is that there will be as many authorized_keys file as the number of servers in a domain (if a SSH access is needed on all servers). A side effect is that the path, mode and owner of the directory which will host the authorized_keys file may be better managed and hardened than before (even if SSH already check those things for sane defaults). It is particularly of interest when managing thousands of servers in heterogeneous UNIX environment, when Solaris, AIX, Linux and HP-UX doesn't have the same ownership same system paths (such as /var, for example).