Without pointing fingers at a specific app, there are several in particular that we have worked with who have "pretty URLs" for their website (e.g. "http://example.com/the-name-of-the-page") but their deeplinks are "ugly" (e.g. "example://page?id=123456"). There is no unified way here to derive the deeplink from the web URL or vice versa. As such, we cannot treat deeplinks and web URLs as always templatized; these are simply URIs whose identities should be considered arbitrary.

On another slightly relevant note, from the perspective of linked data and the semantic web, the <link> tag or "Link" header (both which are suggested by Google and URX) is meant precisely for this kind of reason: being able to link separate resources with a given relation (e.g. "alternate") who may have varying display/device/media qualities (e.g. media queries). Because semantic linkage is between two nodes of the web graph rather than DNS (which I guess would be a subgraph of the web?), one is able to use semantic metadata such as http://schema.org/docs/actions.html to describe not only how to access/consume a resource, but also alternate ways to do so (e.g. from an Android device instead of a desktop client).

Again the above isn't completely impossible to do with DNS, I'll give you that, but it simply is less tenable to assume template-like qualities of all URIs or to ignore the semantic implications of redefining how linked data has worked for the last decades.