I've had good luck using MySQL as the data source of record (setup in a NoSQL style InnoDB table) and then having a Membase cluster which supports replication between different servers. I will probably start using the MySQL/memcache features to play around with them but I don't really see a compelling reason to do so considering that I probably execute less than 10 queries a day against the MySQL version of our caches. I use deflate to decrease the size of my caches so they only really use maybe 2 GB of memory.