When rendering e.g. an icon, we use LibraryFileAlias.getURL. This in turn uses LibraryFileAlias.http_url, which calls LibrarianClient.getURLForAlias.
There is a subtle problem between the latter two. You'd expect http_url to ask getURLForAlias for "the URL for this LibraryFileAlias." But instead, it asks for "the URL for the LibraryFileAlias with this ID." So there's a redundant lookup for "the LibraryFileAlias with the same ID as this LibraryFileAlias." You don't normally notice this of course, because the right object is already cached.
But there is a case where you do notice. If the LibraryFileAlias was retrieved from the slave store but the current request uses the master store by default. Then you do get an extra database query, and on our precious master store—and it also bypasses any prejoining you might have done to get the slave-side LibraryFileAliases in the first place.
The solution: make FileDownloadClient._getPathForAlias accept a LibraryFileAlias as an alternative to a LibraryFileAlias.id (and so skip the lookup). Document FileDownloadCLient.getURLForAlias as accepting either as well. Then make LibraryFileAlias.http_url return self.client.getURLForAlias(self) instead of self.client.getURLForAlias(self.id).
Fixed in stable r11526 <http://