According to this page, Confluence's tiny links are generated by encoding the content ID of a page. Because this encoding is reversible, a page's content ID can be found directly from the link by Confluence without a database lookup.

Unfortunately, the (fairly common) scenario of migrating from self-hosted Confluence to Confluence OnDemand requires space imports, which means all content gets new IDs. This invalidates the 'tiny' part of all the tiny links someone may have used in other systems to link to their wiki content. If they had used non-tiny links, they could simply change the base URL part of the links, but now even that will not be sufficient to get the old links working again.

The suggestion is to provide an alternative to replace the tiny link ID during a space import.