Dedup works on HTTP traffic, yes. Hence the data is kept in a combination of RAM and Disk.
We have also implemented a HTTP file caching feature that will be available for beta testing in a couple of weeks. Since there is overlap between dedup and file caching, not every case will see a perceivable difference. The object cache is useful though to keep some data out of the dedup datastore, which means the memory and disk space can be used for more new data and ultimately lead to a larger datastore of the repetitive data closer to the user and hence better performance in terms of reduction rates and throughput.