Extendible hashing: why does anyone use most significant bits?
I finally went back to the original source paper by Fagin, et. al. They address this:
"We note that if we had used suffixes of pseudokeys instead of prefixes, then the algorithm for doubling the directory would be especially easy: it would essentially consist of making a second copy of the nonheader portion of the directory, immediately after the first copy. However, we chose to use prefixes for the sake of intuitive simplicity (thus, by using prefixes the keys can easily be accessed in pseudokey order, rather than in inverted pseudokey order). "
I don't understand why they saw this approach as more intuitive, as you could dispense with the whole bit idea and go with modular arithmetic instead, but it appears that this was at least their rationale.