Hi,
After nearly 50 posts, perhaps it's time to reprise the design targets and current proposals (preferably with an up-to-date circuit diagram). If I've followed the thread (please correct me if I'm wrong), simplicity and power efficiency are particular aims, the power supply is to be 4 x Primary (non-rechargeable) cells, the output transistor is a "Darlington (pair)" and the two motors are to be reversed with a DPDT (mechanical) switch. Note that if a diode is connected literally "across the motor terminals" then it can only rotate in one direction.
To make even moderately good use of the the energy in alkaline cells (don't even think about the "heavy duty" carbon types for 1 Amp) then the circuit should work down to 1.25 volt per cell, and preferably to 1.0 volt. That "safety" diode in the supply rail is unlikely to drop less than 750mV and similarly the Darlington transistor, so the motor(s) may not see much more than 3 volts. Also, Darlingtons tend to make poor high-speed switches (because there is usually no mechanism to draw charge out of their base) so you might find nearly as much power being dissipated in the semiconductors as in the motor(s). With so many unknowns (e.g. the PWM frequency and its effect on dissipation and acoustic noise, etc.) a "paper" design is probably impossible, and breadboarding essential. However, I'd proceed as follows:
With partly discharged cells, the zener is not going to "see" 5.1 volts so won't regulate (when it's potentially most needed). In any case, 5.1 volt zeners have a poor "knee" compared with the original 9.1 volt device, and zeners below 5.1 are even worse. So I'd just throw out the zener and increase the value of the decoupling capacitor (which now only needs to be 10 volt rated, of course). Also, it's probable that the power transistor can survive reversed battery connections (but the 555 will not) so I'd move the series diode next to the filter resistor (R1), where it won't dissipate the motor current and only needs to be a 1N 4148 type.
Then measure a breadboard at various PWM frequencies to determine which (if any) give acceptable results.
Cheers, Alan.