Fair enough, I'll bite. Given 2^32(or 4 billion) IPv4 addresses, and say, 4 million IP's allocated to the average ISP (I'm being generous here) there's your 0.1%. The rest of the space will go unused since we're using 32 bits to identify these sparse blocks in the v4->v6 translation. Not counting customer /56s, 48s, /60s or whatever.
But customers, and /56s are the essential things to count. Again, you just throw in the number 4 million without explaining where it comes from. IPv4 ISPs come in all sizes with one /24 allocation and some with many allocations of sizes ranging from /17 to /12. Counting IP addresses in IPv6 makes no sense. The addressing hierarchy of IPv6 is designed to have large blocks of unused and unusable addresses. This is both to allow for expansion without changing the network architecture, and to allow for automated address assignment functions which rely on large sparse number spaces to avoid collisions. --Michael Dillon