lat,lon doesn't have any precision attached to it. Also, you might be surprised how many times people get them reversed. Hey, even geojson encodes it as lon,lat.
Also easily normalizing (lat,lon)s isn't so straight forward.
And then there's subtleties when people are actually not using the WGS84 but a crs/datum one shifted a few tens of miles.
Sort of surprised that nobody compares this to geohash[0], though.
I hope that Google will be throwing its weight behind it (seems so for now) and proposes a rfc for this so we can stop bikeshedding about geo encoding standards.
Geohash is a great idea, but at the end of the day all these systems are a mapping to latitude,longitude. The lat,lon precision can be derived from the number of digits after the decimal point.
The main argument for the existence of these systems is that lat,lon is hard to remember. Here is a simple solution, convert any lat,lon pair to an easier to remember pair (ie a pair where numbers are repetitive, consecutive, etc).
for eg: 45.89988,-64.36288 -> 45.9,-64.363
and so on. It is a simple hack making ll memorable without sacrificing much accuracy.
I agree with the bike shedding comment! It seems to me that the location format is the most trivial part of an addressing infrastructure. Any system could easily be successful, but it needs a level of infrastructure and support to work which is not trivial. My guess is that the eventual solution will be for apps to use GPS and camera to identify delivery points in places that lack good addresss infrastructure.