I've worked a little bit on a real time system for tracking vehicles on a map. We did not use any 3D stuff in it. The whole path and weight data was on another tracking system (It has dynamic weight on the paths due to traffic and road blocks etc.). We pulled the real time data from there and just adjusted cropped image maps of the place so we had the references all set. We loaded at least 8 map units to surround the zoomed area and make zooming faster.
The path algorithms used were 9 custom and 2 standar ones to check for errors etc. ( the ones you can see in books like Dijkstra, etc.).
One of the hard parts was getting the roads to work like roads ( Meaning there's road data rather than trying to get gps position over the map) and mixing it up with terrain and whatever other entities you have there.
We managed the memory by cropping a map (square area) and adding road data (array of nodes array to get road directions and finding where roads meet and with coord data) and serializing that whole thing as a map unit and we rendered that unit and the surrounding ones (
.
We had different resolutions for the maps because sometimes it was needed to zoom a huge area so we built same size maps with much more area and data in it and we serialized again. So when you zoomed out too much most map data was already in place all was neede was to load more data from the other system.
Of course this was a huge machine running it. And the rendering was done apart from the processing.
I really don't see any benefit for using jPCT here thought.
And you do need to get some good maps of the area you will use and work it from there.