We wanted to do XRP calculations internally in a 64-bit unsigned with one bit reserved to signal that it was an XRP amount and one bit free as a sign bit. To avoid having to deal with overflows inside operations and simplify the code, we didn't want to use more than 1/16th of the legal range. For human convenience, we wanted the main unit to be divisible into millionths. That set an upper limit of 2^(64-2) / (16 * 10^6) or 288 billion. 100 billion seemed the most human-friendly number within that range.