We're using Polymer.Base.importHref to load parts of our application when they become needed.
This works quite well, except when the network environment is flaky: if the first importHref for a specific page fails, then it will never start working for the same URI until the whole application is reloaded.
From what I understand this might be intentional: importHref creates a link element dynamically, and the browser should not load the same document twice when using import links.