These aren't restricted to the old 7 bit ASCII character sets though, are they?

We've already had cases where people were registering famous domains with one character changed so some high unicode value that was graphically but not conceptually equal to the original character. Net result, new domain whose address looks and reads identically (to a human at least) to the domain it's trying to impersonate. Hopefully there are safeguards in place but that happening at a TLD level would hardly be pleasant.