There is the problem, that human-like robots are not yet able to move as reliably as humans. Not far ago was there content for robots to get out of car, walk to doors, open the handle and walk inside - a lot of participants failed miserably at such "simple" work.
There is so much possibilities for movements, lot of inertia, not well defined ground and not perfect visual recognition in real time.
On the other hand machine-like robots can move on rails, have perfect information about its position, only few axes, all programmed for optimal movement and stable envirnment. They are able to do limited number of action much faster and with a lot higher precision, than humans can.
So for ongoing couple of years the non-human robots would be used more, as it is much simple to get much better result from them.
The only field, where human looking robots can be better is interraction with people, for psychological reasons. But while from X-Y-Z laser you expect getting some king of high-tech program and then cut material fast and effectively, doing many times the same movements, form human looking robot you would expect human-like behavior, mainly ability to recognize gestures and face-expression, native language and vague defined commands like "I would like something sweet, but without too much suggar and rather mildly flavoured as I am on the diet" - which is a too much more difficult task to respond as expected from human like inteligent being.
And the reconfigurability is higher if you can just change some robotic arm ot drilling tool, in contrast to difficulty to teach effectively manage tool designed for human hands (which cannot be hold as firm and in perfect position, as something, where you screw two linear plates together to make standardized connection and wire the motor to the driver, instead of pressing some oddly shaped switch with hand full of servos.
