Exactly. And the *reason* why IPv6 has 128 bit addresses is because
the designers realized that such losses happen, and ruled out 64-bit
addresses because of that effect.
Since those losses are not significantly diminished by doubling the
address length, why bother?
The problem arises when zones of the address are reserved. Setting
aside n bits of an m-bit address diminishes the address space by 2^n,
not by (m-n)/m.
I suggest you figure out just how much bigger 2^65 is than the
current 2^32 ...
You're making the same mistake as everyone else: You calculate the size
of the address space as 2^n, but you believe that reserving m bits
diminishes the address space by only m.