https://en.wikipedia.org/wiki/USB#USB_Power_Delivery_specifi...
I suspect that the 100W figure is actually just a pulsed maximum specification. The thing is that current ratings for wires are actually specified as a max. continuous current for a given temperature rise in the conductor, per unit length. So blowing 3-4x the current through a conductor for a very short time (think of flashing a bulb or moving a servo) is not a big deal. You just get a transient heat rise. Also, most applications simply don't require 20A. Even microwaves and kettles stay below the 15A residential fuses. (Although they do come close. I once had a shitty basement suite with an underrated fuse. If I ran my toaster and my kettle at the same time the breaker would flip!)
It should also be noted that a switched mode voltage converter can have well over 90% efficiency, even with large changes in voltage.
You should also remember that those microwaves and kettles are getting up to 15A @ 120VRMS continuously, which works out to 1800W. You can verify the actual power output of a kettle by timing how long it takes to boil a liter of water, and calculate power from this time and the specific heat capacity of water.
In any case, as dfox mentions, most internal voltage level conversions won't be switched, because it adds complexity. They will be some form of linear regulation s.t. they can move between logic levels. That's different than moving from whatever high voltage is on your 150W USB line into a level that won't fry CMOS circuitry. There's a reason that the wall-->DC plug conversion usually happens in a brick on your power cable. Switched mode will be used as sparingly as possible, such as when you also need AC signals rectified, if you need both buck and boost depending on a battery or something, or if you need to be able to modify the control loop dynamically.
> those microwaves and kettles are getting up to 15A @ 120VRMS continuously, which works out to 1800W
Well that's kind of my point. Even at 1.8kW those devices don't need to draw 20A continuous (or even pulsed, because of the fuse). Basically no matter what you're doing, the copper losses are roughly fixed by the hardware. What you can control are heat dissipation and current levels, and it's a lot more fun to play with Ohm's law than try to fight against thermodynamics.
http://circuitcalculator.com/wordpress/2007/09/20/wire-param...
I've had USB cables get hot before. The finest being a cheap Chinese USB cup warmer. Why they build such things escapes me.
The losses at low voltages are pretty high as well.
There's a reason that we have mains cables how they are and our current power distribution infrastructure.