I do wish these platforms were a little more agnostic to the technology used in their API, and had simple C bindings anyone could write glue code for to use their favourite language. Learning yet another markup language and using C++ in order to create apps doesn't really appeal. Whatever the platform insists on as glue is what we'll have to use though, as usual. Personally I'm going to look into the webapp support in Ubuntu Phone to see if it supports local web apps, and hope that I can use that glue to get at any device state that I need to access.
As to qml in the browser, I'd really rather have web apps on the desktop than start writing websites in qml, I'd be very surprised if that catches on.
How does this get around sandboxing on iOS? My fuzzy understanding is that components would need to be downloaded for each app independently. And is there a reason those shared libraries can't be bundled?
According to the qt blog, only qtquick2 depends on v8.
It is pretty close to write once, run anywhere. The only downside is if your app is the first qt app installed on an ios / android device, it forces a download of the shared libraries which are quite large. You also have to build the applications against individual platforms, since you are using C++, and the finished binary is a native application that uses the qt libraries to launch the qml engine and v8 for the javascript parts.
One experiment I am anxious to see come to fruition in the qt space is the KDE project to get qml apps running in the browser : https://projects.kde.org/projects/playground/www/qmlweb. Since QML is inherently javascript, a web app could only use javascript / qml to create a restrcited sandbox (or nacl / other browser equivalent for the c++ parts) and the entirety of qtquick is designed around high latency elements, so it can support network transport.