> It would also mean every single word entry would have to hit the root level of this public name lookup system since the client wouldn't know until after it checked.
Caching at multiple levels should work fairly well for this?
Theoretically but e.g. google.com has a TTL of 5 minutes, TLD responses usually have something like a 48 hour TTL associated. Drop that cache timer then multiply by a huge factor because everyone is hitting TLD lookups for random things like "sausage" instead of ".com" and it's easy to imagine the majority of the root traffic volumes becomes bogus lookups instead of actual requests. Once it's into ".com" it's off the public infrastructure and volume becomes a problem of the current pay to operate TLD system.
Could that be done right? Sure, at which point Google would probably want to keep the 5 minute TTL instead.
Caching at multiple levels should work fairly well for this?