We use index mappings at TaskRabbit, and it does make things a lot easier. It helps with 0-downtime mapping changes (as you point out), and it also lets us combine large indexes. We segment large data collections by time (ie: events_year_month) which we roll up in an alias. This allows us to delete/add large collections without downtime as well.

What's cool about the elsasticsearch api (and therefore the elasticdump tool) is that with aliases, you can read/write to and from an index just like you can an alias.