Instead of upscaling, one instead compresses a higher resolution version of the video.
Oh, this is why the other patent describes an upscaler. It produces a higher res version of the input frame, which then is fed to the compressor. That would explain why the VDK was noticeably slower when enabling the zoom option, because the upscaler does its own block searching too.
I guess it explains why the zoomed videos did not look so great. The upscaler produces artifacts, then the compressor produces more on top. Some things did look sharper, but when my boss tried normal photos, we did not see any compelling difference from traditional upscaling. Which is understandable, even if it used fractal interpolation (which it does not), you do not see new details. E.g. skin still looks flat, you do not see pores when zooming in.