Nifty work.
> Even though precise results are returned by the plate tectonic models, the locations are still approximate [...]. In my tests I found that model results can vary significantly.
One challenge with science education content is to avoid nurturing rich ecologies of misconceptions... to the degree possible given resource constraints and conflicting objectives. So here are some random brainstormy ideas.
Add small inserts with different plate models? To reduce the implied certainty. Uncertainty is usually handled poorly in science education, with unfortunate results for education and society. Perhaps pixelate the globe?
The false colors represent elevation (EDIT: no, Scotese's hand-drawn sort-of-elevation maps), not biome. So "green map" may represent a long-term tan desert. Perhaps choose a less pretty but 'less likely to be misinterpreted as biomes' colorization? Maybe as an option? A biome map would be great to have... but at least as of some years ago, it didn't seem anyone was attempting unified paleobiome map models. It seemed more lack of incentive than lack of data, at least for some periods.
Sea level and coastline position varies at high frequency. 12k years ago, the shortest walk from Boston to the sea, was due south past Long Island. A color scheme that deemphasized the coast would allow coastline variation to be averaged/blurred.
Cloud coverage and patters vary greatly across time. Pretty clouds over Pangean desert made me smile. Perhaps explicitly call them "Bogus Clouds"? :) Science education content frequently shows bogus things. Rather than difficult reform, simply becoming more transparent/honest and comfortable with calling them bogus might be a relatively easy but significant cultural improvement.
The star field is missing long-term features, like a visible galactic disk. Having a higher-density star map might also usefully obscure that the set of bright stars is changing. Or perhaps just another "Bogus" label? They're really quite handy. :)
If anyone has time on their hands, it might be neat to attempt a more principled representation. Perhaps realistic appearance, averaged over a few My? Average cloud cover maps are less pretty than weather, but plausible long-term ones might be derived from relatively simple paleoclimate models. Similarly for biomes.
Paleoglobes are traditionally very misleading. As we move forward, it could be nice to improve on that.