What to do about single-word domains? Is "google", standalone, a valid domain? This is a big problem for browsers which use the same input form box for URLs and search keywords. Do you look up in DNS first, or with the search engine first? DNS no-finds are relatively expensive operations; they may have to go to the root servers to get a definitive "not found". Having a relatively small set of TLDs avoided confusion between domain names and search queries.
Then there's the relative domain problem. Although it's not used much any more, it's a feature of DNS that if you're on machine "abc.example.com", and you look up domain "xyz", it will try "xyz.example.com". The idea was that you could get at servers within your own organization without typing the fully qualified domain name. If you want the domain to be searched from the root, you're supposed to type "." after the domain, as in "example.com." Nobody knows this unless they're really into DNS. There's an adspam exploit that uses relative lookup to turn certain mistyped domains into ads on the domain "com.com", which pretends to host all subdomains of the form "*.com.com". If you're on "example.com", and you look up "nonexistentdomain.com", which doesn't exist, the domain "nonexistentdomain.com.com" is tried as a relative lookup. You get ads.
Then there's the problem of determining what's a second level domain. For that, you now need a very big list. (The list is here: https://publicsuffix.org/)
Current thinking seems to be to disallow single-word domains. There are already, though, a few two letter country code domains which host valid web sites at the TLD. (Try "http://ca.")
This is already a problem for intranet, e.g., if there's a web server accessible by just the machine hostname. Most browsers default it to search unless you remember to prefix it with "http://" every time (even if "devbox" is already in your history... thanks browser guys). It makes me miss the days when the search and address bar were totally separate.
Then there's the relative domain problem. Although it's not used much any more, it's a feature of DNS that if you're on machine "abc.example.com", and you look up domain "xyz", it will try "xyz.example.com". The idea was that you could get at servers within your own organization without typing the fully qualified domain name. If you want the domain to be searched from the root, you're supposed to type "." after the domain, as in "example.com." Nobody knows this unless they're really into DNS. There's an adspam exploit that uses relative lookup to turn certain mistyped domains into ads on the domain "com.com", which pretends to host all subdomains of the form "*.com.com". If you're on "example.com", and you look up "nonexistentdomain.com", which doesn't exist, the domain "nonexistentdomain.com.com" is tried as a relative lookup. You get ads.
Then there's the problem of determining what's a second level domain. For that, you now need a very big list. (The list is here: https://publicsuffix.org/)
Current thinking seems to be to disallow single-word domains. There are already, though, a few two letter country code domains which host valid web sites at the TLD. (Try "http://ca.")