Another fairly simple scheme is the current limiter below. It will only drop about 300mV @ 0.5A, but the MOSFET will get really hot under sustained 1.5A load.
I calculated that the motor resistance is 24V/3.5A ~ 7 ohms. This means that, at 1.5 amps stall current, it will drop 10.5V, which puts (24-10.5)=13.5V across the current limiter. The 0.5 ohm resistor drops 0.75V, which leaves 12.75V across the MOSFET. 12.75V*1.5A ~19W. If you don't anticipate high current loads except during start-up, you won't need a really big heatsink. However, the MOSFET would then burn up if the motor did stall.
Does anyone know what the saturation voltage of an LM317 (or LM338) is in a current limiter configuration, when the load won't support the limit? I suspect that it is close to a volt, but I don't know. That would be in addition to the drop across the current-limiting resistor, which would be about 0.85 ohms.
I like Steve's resistor/switch combo. If you had a centrifugal switch, that would be really simple.
None of these schemes allows maximum starting torque.
EDIT: Added a zener from gate to GND.