I use vector layers on top of either a blank basemap or an aerial photo. Users can turn off my vector layers and use a real esri basemap if they want. I generate a layer for labels and one for the basic stuff (roads,trails,water,parks)
This has worked quite well for us. Currently there is a state-provided aerial but the Bing aerial is good or better. So users can turn on either the aerials we provide (city and state) or Bing as the basemap and our vector layers look good on top of those. If they opt for a full basemap then they can turn off our vector layers.
By default no aerial layers are turned on so things load fast and people get a good looking map even if they don't know how to turn on aerials or basemaps.
If you were generating only official esri basemaps then you might want one that's vectors only and one that has a built-in aerial. Since aerials have to be rasters then you would need raster tiles.
You indicate 0 tiles in the raster, that just means the layer is set to cache tiles on the fly. The cache will grow over time as people reference tiles and they get rendered. This makes it slow at first but speeds up as areas of interest get cached versus generating all the tiles in advance, which can take hours to days and uses lots of space.
Personally I found using vectors so fast and maintaining caches of raster tiles such a pain that I quit that. We're either feature services, map image layers, or vectors now.
You can see what I am talking about at https://delta.co.clatsop.or.us/apps/ClatsopCounty/
The vector layers are "Clatsop County Labels" and "Clatsop County"
