Part of it is imports. The first sidewalk import happened in Heidelberg, Germany, but it focused on sidewalk=* tags for consistency with existing tagging in the area. Within about a year, various members of the U.S. community began a wave of sidewalk imports that added ways, starting with Seattle; Washington Court House, Ohio; and San José, California. The San José import alone added some 3,409 miles (5 486 km) of sidewalks, which we manually connected with more than 175 miles (282 km) of crosswalks.
(Ironically, for all the angst we saw about sidewalk ways, we actually had to do nontrivial work to have linear geometries, because the City of San José only provided polygons. Who knows, maybe we’ll go back and add in all the area:highway=footways we’ve been sitting on for the last eight years. We still have an opportunity to outpace Google Maps right in their backyard. So far they only show sidewalk areas in a few smaller cities like San Francisco. 
)
Regardless, as these charts show, there’s been a lot of organic growth too. It’s only natural to want to draw something you can plainly see on the map. Tags on the roadway are apparently less intuitive to many mappers.
By the way, the query I shared distinguished sidewalk=separate/no and sidewalk:left/right/both=separate/no from other values. I think this should account for mappers who only use these keys to indicate the existence or nonexistence of separate ways. sidewalk=lane is one of the proposed tags for pedestrian lanes, but those are relatively rare in the U.S. anyways.