You're saying that Alice who sold 99% stake to Bob would have her chain outcompeted by someone transacting with .5% share on top of Bob's chain, presumably because Bob would wait for confirmations before accepting the transaction, while others are building on top of it.
But what about when Alice owns 30% stake, sells it, and secretly has another 30% stake waiting in the wings.
The fact is you don't know how much stake an attacker has, (how much of their stake is going into the attack).
30% stake loses with 70% stake. That's the whole point.
True chain confirmations example:
Alice's 30% -> same stake, buyer's 30% -> then someone (many someones) else with 35%, in effect
30%+35%, 30%+35%, 35%
later, even more stake.
False chain:
60%... nope, 60% < 65%, rejected.
Now there may be a problem in practice, in that if there's not enough stake active it may take some time to destroy the illicit fork with big amounts.
However, owners of the currency know it, so they should either run a full node for this very reason, or lend their forging to someone they trust. It's their money after all.
Zipf distribution seems to describe the wealth concentration best, so in fact it's practical.
Even if you're pessimistic and say that's 20% of the coin is enough to fork a coin for several hours, that's still much larger security than in PoW! (as % of market cap).
The amount of full nodes in bitcoin doesn't mean anything here, as there's no comparable reason for bitcoin holders to run full nodes.