There was no Cell based GPU - there was however a Toshiba project designing the PS3 gpu before NVidia got involved (https://www.neogaf.com/threads/ps3-cell-made-sonys-first-party-what-they-are-today.1636357/) there's some commentary on what might have been in this thread.
Tons of stuff - higher-fidelity realtime lighting & shadowing systems, bigger NPC crowds, substantially more complex AI interactions (and more physics interactivity) etc. It's not entirely dissimilar from MGS4 example. The resolution was always the same too - the only thing that changed in retail game was upgrading the scan-out buffer to full height (so when frames are dropped, you don't drop res) but rendering was already @ full frame.
I gave very concrete examples of exactly that. Both Division and Watch Dogs were little more than concepts when first shown - that entire 'city block' they demo-ed was little more than a thinly disguised corridor of near-field detail + clever camera work to make it 'look' like going around a city, and the gameplay - while 'playable' was almost entirely canned. That's what 'vertical-slice' demos most commonly do - you concept out blocks of what you want your game to look & feel like - you don't actually build all the layers out in cut-down fashion (that'd cost orders of magnitude more to do and would never be on time).
And doing it in realtime has certain benefits - not the least being confirming it's at least somewhat plausible to run on target (that is - assuming that your target isn't completely made-up, something all these 3 examples had in common).
Indeed - this is already well into 'hw iterations became familiar' territory, they knew really well by then what they were doing on PS3 and what was realistic.
