I am setting up a load-balanced cluster to host a PHP application with
several machines that will be running the same code. I was thinking of
putting most of the code itself on an NFS share, and counting on APC to
cache it in each server's memory so the share would (theoretically) be
hit very little. Is this a sound idea? I know that APC uses a device
ID and inode to identify cached code; is this a problem with NFS?

The production code will not be changed very often once it is in place;
I figure I can disable apc.stat to avoid hitting the NFS, and just flush
the cache by restarting apache on all the servers on the rare occasion
that the code is updated. I also plan to give APC copious amounts of
memory and keep apc.ttl at 0 so that code is cached indefinitely.

Can anyone who is intimately familiar with APC tell me whether this is a
sound idea - or would it be smarter to keep local copies of the code on
each server?

Thanks,
Jeremy