My analysis of roadmap code is that it does not load whole county files into memory all the time. It uses memory-mapped files and reads what it needs, and the author already has a geospatial indexing scheme- that's what the 'usdir.rdm' file is for. So we already have a pretty decent indexing scheme going. I also don't think file size should be an issue on the empeg. If ths issue is File size vs CPU then CPU should win over file size every time. The one thing the empeg has going for it is it has more storage space than your typical embedded machine.