I think this and bug 15657 can be resolved as part of the following premise:

When an element is in template contents, its properties are detached from the internal state of the element, and instead simply simulate the attribute accessors.

For example, "liveness" of an image only comes into play when you set its "src" property. If the property simply parrots setAttribute/getAttribute and doesn't change internal state of the element, the effect is that the image is never fetched.

Similarly, the tolerance of setting "tabindex" is not significant, since the property is not actually set, and is instead using setAttribute/getAttribute.

We could do this by:

1) defining that template contents are parsed as a document fragment that was created from a document without a browser context (var df = document.implementation.createHTMLDocument('title').createDocumentFragment();)
2) modifying HTML spec to clarify what HTML elements with a ownerDocument that does not have a browser context should do.

At the moment, the interoperability story is pretty bad here, with Gecko/IE9/WebKit doing slightly different things, Gecko doing the thing that's the closest to desired behavior.

That's browsing context, not "browser context": http://www.whatwg.org/specs/web-apps/current-work/multipage/browsers.html#browsing-context

*** Bug 15657 has been marked as a duplicate of this bug. ***

Added the notion of "inert" in http://dvcs.w3.org/hg/webcomponents/rev/cb6f77f5a6bb.

template contents initialization algorithm takes care of this: http://dvcs.w3.org/hg/webcomponents/rev/195ce90ed523