The short answer:
It's harmless. This is the correct behavior when loading translated resources. To avoid the multiple hits and 404s, you need to do a build.
The long answer:
When loading translated or locale-specific material, the default mechanism is to conduct a search for the resources best matching your locale by making successive URL hits along a search path. For English, United States variant, Dojo would first look in the 'en-us' nls directory, then in 'en', then in the root for the fallback. The results would be mixed together with the most-specific ones overriding the successors, such that only the modified strings in each variant need to be present. Very often, a variant will have no content at all, so a 404 will occur.
It is not possible to predict the existence of these URLs at runtime unless special pre-processing occurs. Dojo's build process is able to flatten resources and combine them into single locale-specific files such that a single hit will fetch all of the needed code, and a second hit will fetch all translated resources. This will avoid the 404s and also optimize loading by avoiding the network latency of individual web hits.
