There is, of course, no way that an 80% power consumption can be achieved just by reducing seeks. The obvious ways that it might be achievable surely includes using fewer, larger and slower spinning hard drives as the assumption will be that the flash cache (horrible near rhyme) will handle the considerable majority of IOPs. Powering down the drive is surely a non-starter for all but semi-archival type storage as it takes many seconds to spin a disk up to speed which will mean the requestor is stalled for that long. Given that spinning up a disk consumes more power in the short term it would probably only yield a net power saving if the intervals between access were measured in minutes.
As with all cached solutions, then it's success will depend on the locality of reference of the access pattern. We know pretty well from enterprise storage arrays what the effectiveness of a multi-GB NV cache is in fronting HDDs (which depends greatly on access patterns). What Flash does offer compared to the battery-backed up NV cache in an Enterprise array is a considerable decrease in the costs per GB of cache, but then again, it isn't going to perform as well. Of course SUN have already gone down this route at an array, not device level.