Lurker since 1997 and occasional WikiGnome.
I am a believer in TransparentPersistence, and see PrevalenceLayers as an effective low-end solution to the problem. I've written one or two to support oddball platforms; in my current work it's rare for the data set to exceed 5 MB in RAM, and usually produces snapshots under 20 kB on disk.
The biggest complaints I get are that a number of other programmers just plain don't like designs based on LiberatingConstraints (for example, PrevalenceLayer requires SeparateIoFromCalculation), claiming that the use of design constraints breaks TheSimplestThingThatCouldPossiblyWork for new requirements.