QML is not inherently slow - but I do agree that at this point it's all too easy to get carried away and write things in ways that do incur a performance overhead. It really helps if you know how the underlying tech 'thinks' (this is not QML specific, writing C code anticipating how the assembly will look like also can give very interesting results). In this case, however, it's about our presumption that everything hardware accelerated must be faster than anything done on the CPU, but reality can get in the way (I think the most famous one in Maemo land is that the N800 does media playback better with *lower* clocked DSP). The OpenGL situation is a similar one - there are many caveats that can (and do) kill OpenGL performance so on the N900 you're often better off relying on the relatively zippy CPU.