Question about mobilenet

Is using MobileNet always better than normal conv networks?

Professor said mobilenet reduces computation and make the model has richer functions. So i thought that if we use depthwise separable convolution to the neural network we studied before, the precision of the models increase??

Thank you.