No, I'm not sure.
I don't recall the details but the last time I upgraded storage and moved the contents I did see evidence of Java in the file system.
I can't imagine why they would use Java for some things and not the UI. JavaFX fits. It's an obvious platform for loadable apps too.
As for hardware graphic acceleration... depends on hardware capability at a minimum. I doubt they would tune for each box; probably using the lowest common denominator method across all boxes so it will be very low-end. Note that new iOS releases cause obsolete hardware a lot faster than DTV does.
For clients, RVU sends "bitmaps" (compressed one would hope) to the client for the UI. There isn't much hardware acceleration possible with dumb data like that. I doubt RVU is any better than VNC or the like which isn't saying much (heck, X11 used to be far better than VNC).
As far as remote display of polished UI transitions (like crossfade in and out of menu or guide)... I can get Aero and GUI effects with UltraVNC and or Dameware mirror drivers. Or Microsofts new RemoteFX in Server 2008 R2 and later.
It would be nice to see things cross fade in and out like Passport guide on Cisco set tops. Or things like the banner slide down and slide off when invoked / dismissed. Or a caller ID notification slide into the screen from the lower right, then briskly slide off screen when dismissed. Or the mini guide slide up when initiated and slide down out of view when exited. It just masks loading. Like the CPU can be preparing the next screen while the video is "transitioning" to that screen. Makes it feel more responsive because it starts to enact a visual change at the press of a button. Video processing would have to be double or triple buffered. I know Java can execute such things as the Xfinity X1 platform is based on Java in the Tru2Way specification. It's running Intel CE4100 CPU, however their Pace RNG150 HD set tops are the usual Broadcom based variety which state "accelerated 3D graphics" in their spec sheets.
Quite honestly I don't get why providers don't move to the ARM platform. A good 1.5 Ghz quad core ARM has more oomph than any cable or satellite providers equipment. Plus they can dynamically disable unused cores and go into ultra low power states. I can decode H.264 HD video in the palm of my hand on my iPhone or my wife's Samsung Galaxy Nexus. Yet that is battery powered device connected to a WAN, wifi and Bluetooth all running, alongside listening for phone calls, txt, push email, etc. Just amazes me what a difference in computing power a $200 + 2 year contract cell phone has compared to again a few hundred dollar + 2 year contract DirecTV (or competitor) DVR.