Has anyone ever considered witting some sort of interface to register a function to so that newly stored keys or values are subsequently processed in parallel in the background with the output being used to either produce a Tag (as you are proposing to add) or a replacement value (with perhaps a flag to indicate that it's been completed)? I realise this is significantly adding to the basic functionality and hence base purpose of memcache to the same extent as the memcachedb for example so it might be a completely seperate fork or mod_func style effort but was just curious? If it were a "proper" (not entirely sure what I mean by that as yet) parallel processing method you'd have to add some sort of processing scheduler either to each node or to a designated master node that would hence need to communicate with/manage/know about/ monitor the other nodes as a first step.
Regards, Philip Mather
