Scale definitely makes a difference, although I think it's complicated by the resolution of the preparation. From my standpoint, just-in-time creation lets a DM run a much (geographically) larger sandbox without the prohibitive levels of advanced prep required to create sufficient detail to run the game in such a large sandbox. Instead, the early prep can be all high-level, with the time spent making details on the fly limited to only those areas that ultimately show up in the campaign.
For illustration/comparison, the last sandbox campaign I prepared covered a substantial portion of a continent. There were dozens of major items of interest (e.g. cities, settlements, named geographical features) on the full map that the characters had (although that map was IC known to be outdated) and dozens more in my notes. Given that the areas of interest were days or weeks of travel from each other, each would have been surrounded by its own set of hundreds of hexes (at one-hex-per-day scale).
But my total up-front prep was probably less involved than what you're describing, as each point of interest only had a couple sentences in my notes, or a bulleted list. If/when the PCs actually decided to head for any of these points, that's when the just-in-time creation would take place, expanding my high-level notes into low-level detail. Based on previous campaigns, only a small fraction of the high-level areas of interest would actually be visited, so this method avoids wasting low-level prep time on areas that are never seen up-close.
It sounds like in your game, a much higher percentage of the original map will likely be visited, so maybe the need to avoid wasted prep is less pressing? What fraction of the map do your PCs typically visit?
Also, I never use random tables for content generation, so just-in-time creation lets me limit myself to what is in a given location now instead of spending even more time in advance having to figure out a schedule of what would be in that location at any arbitrary time.