Assuming you have 3 load-balanced servers configured (server_a, server_b and server_c), each running a local memcached server and the memcached configuration is:
Enable clustered servers: Yes
Set Servers: server_a, server_b, server_c (each on one line)
If the memcached instance dies on server_b, then it will silently fall back on using 'file' caching. This means that:
1. Any cache updates written on server_b will not be propagated to the other servers (as the memcached plugin code to update all the servers will not be reached - the memcached plugin will already have bailed out)
2. Any cache updates written on server_a or server_c will not be propagated to server_b.
I'm not sure if there is any solution to this issue, but it could lead to some problematic inconsistencies between the different servers.