Comparing frameworks using the callgrind maps doesn't really say a thing in terms of performance. The relative weigh of parts inside the framework can be visualized, though you can not conclude by any means framework A is slower than B based in these figures.