I was talking to a certain Dutchman (tho not flying) who claimed that a certain system developed at his library was actively harvesting all references from the various providers they used in order to provide “lightning fast” performance (you’re searching in a bibliographic index and you get better performance than you get in the native interface).
Smoke and mirrors. What you’re seeing though is that the price of storage is low, lower than it’s ever been. It makes no sense not to harvest data if this adds some value to your library. Google’s MapReduce provides a way of accessing these enormous quantities of data, at the same time as users getting ever more impatient. Net result? Faster systems requiring on-site storage. Oh, and we’ve got the multiple terabyte storage, and petabyte exists, but is pricey.
My prediction: within five years we will see harvesting becoming the norm for libraries, after all z39.50/SRU and their like aren’t really suited to today’s pace.