Further thoughts on this:
Calling the shuffler "broken" is hyperbolic/click batey. I went with that for thread title because that was title of CGBs video posted in OP. The addendum I included in 1st line of OP is more honest - its exploitable. Of course you can still play games with same decks you'd run in paper and have fun/win, but you can also game the statistical averages with shuffler to your advantage if so inclined.
Here Some Dood has made 13 land decks from all monocolor options:
mono whitemono blackmono bluemono greenAnd we already saw mono red from CGB.
These lists aren't even optimal (he built to his preferred playstyle rather that focusing on the best possible version - for example, in the mono blue vid he talks about running more counterspells with the extra card slots being a Better version of the 13 land deck; he doesn't like that style of play, so he ran other spells instead), but they're proof of concept. They are playable decks. Mono green was pretty weak, and mono red is clearly the best, but the rest of the pack aren't Bad.
Another Dood ran the numbers on over a million games, and shows pretty clear evidence that contrary to the claims made in MTGATracker's
Myth of the Evil Shuffler article, it's not so much a myth as it is a real thing that happens. The data gathered does indicate that if you have X lands in your deck and your opening hand contains Y lands, you have Z% chance of flood/screw. This knowledge allows savvy players to mulligan themselves to better Z% rates.
https://www.reddit.com/r/MagicArena/comments/b21u3n/i_analyzed_shuffling_in_a_million_games/(And this commentary video about the evidence compiled by Another Dood, which I found useful because it linked to the reddit post/source, but couldn't finish the video because I find the voice/commentary style to be a bit irritating in the Angry Pundit sort of way:
Desolator Magic: MTG Arena Shuffler PROVEN To Be Rigged/Malfunctional in Large Test)
It's funny because as mentioned in the OP, I noticed similar trends in Duels. Specifically with land flood/screw based on land count in opening hand on Duels. Like, everyone who played Duels knows if your opening hand has 5 lands, you are guaranteed to draw at least 4 more lands in the first 5 draws (but you can break the pattern with a shuffle effect, like Evolving Wilds). Having 2 land opening hands lead to screw more often than it should - but it wasn't quite as consistent as 5 lands = flood draws.
All in all, I don't view this as game breaking. I personally wouldn't care too much about gaming the % in causal play - but in a highly competitive setting like a tournament, I would def use this information for the competitive advantage. And I still think it's hilarious that you can play 13 land decks in Arena. More novelty than dominance, but hilarious none the less.