Esri recommmends fused caches whenever possible. The difference is that with a multi-layer cache, each layer is cached individually, so you can turn on and off each layer individually. A fused cache takes all the active layers in a map service, 'smooshes' them flat (effectively turning all layers into one), then caches using as many scale levels as you define. You'll likely need more than one scale, otherwise your map will only draw at the one scale that you cached. Can you try with more than one scale set up?