Zerogee
Clencher's Bogleman

There are ways of doing this.. Some better than others..
A very simple way, is to use a string of back-to-back diodes, but they will get warm, and have other limitations as well.
The main problem with 'over-volting' the motor, is it may saturate the windings, causing heating.
At low speeds the PWM '24 volts' will be off more than it is on, so the 'average' voltage is lower and the loco runs slowly. The power (when it is there) is the full 24 volts, so you get a better 'push' to the motor, and better low-speed running.
Without an oscilloscope, you can not easily tell what voltage is coming out of the motor terminals on a decoder. Also, being a 'square-wave' means your normal AC meter wil give an incorrect answer.
That's pretty much how I've always understood it too - that if you're feeding PWM from the decoder to the motor, and the supply is nominally 24 volts, then at half-throttle the power pulses will be half at 24v and half at 0v, so as far as the motor is concerned it will see an average of about 12v..... the question then is whether the continual "spikes" of 24v will hurt the motor at all - my understanding of that has always been "no", and going back to Phil's earlier suggestions my feeling would be to give it a go and see what happens; if the motor gets hot, then you can look at other methods......
Jon.