Mildly off-topic, but I've always wondered about having a non-fixed length address scheme: if there are 64 bits in the address, let the first 8 bits (64/8 = 8) be 1 and the 9th bit be 0. This way we encode the length of the address. After that, every bit is available for use (all 64 - 9 = 55 bits).

Suddenly want to increase the domain from 2^55 to something much higher? Use a 15-byte address where the first 15 bits are 1s and the 16th bit is 0. The rest of the (15*8 - 16 = 104) bits are available.

You get the point - we can arbitrarily increase the size without giving up flexibility.

What's the drawback if we use something like this? Routing troubles? Speed?

Alternative method: the addresses can be as many bits as required, but every bit index that's a multiple of 8 must be 1. The last byte is signified by a byte having 0 at index 0. Basically:

    [1]1010110 [1]0001111 [0]1101011
                           ^ last byte