It doesn't consume all that much CPU, though it is possible to saturate it. The trick is to take advantage of the smooth scaling model: you an add memcached nod

It doesn't consume all that much CPU, though it is possible to saturate it. The trick is to take advantage of the smooth scaling model: you an add memcached nodes until you run out of nodes. The clients have a list of nodes as part of their configuration, and the client runs the hashing function, which includes mapping the entry into the proper node. So the only real overhead that each node has to deal with is serialization. It's very simple, so it works very well. :)