I combined Robert Munafo's stepper motor (http://home.earthlink.net/~mrob/pub/lego/stepper.html) and Rob Limbaugh's partial rotation (http://www.abs-robotics.com/locomotion/partialrotation.htm) to get a working prototype of a base 5 odometer.

Description:
This counter counts in base five up 244 (base five) or 74 (base 10). (It could just as easily count to 444 (base five), except that I ran out of 40t gears.)

How it works:
Large tire drives the mechanism.
White wheel next to the tire moves continuously with the wheel. In a vehicle odometer, it is analogous to the tenths counter.
Center white wheel moves 1/5 of a rotation for each complete rotation of the first wheel. However, the motion of the center wheel is not continuous. The 24t gear which is attached to the first 40t gear only meshes with the 8t gear for about 1/10th of a rotation (of the 40t gear). (Credit this to Rob Limbaugh.) Once the 8t gear next to the cam rotates more than half way around, the small black rubber band pulls it around to a full rotation and keeps it there until the 24t meshes with the 8t again. (Credit this to Robert Munafo.) The motion of the last wheel is similar to that of the center wheel. Because the camshaft is meshed to a 24t instead of a 40t, the last wheel turns 1/3 of a rotation for each complete roation of the center wheel.

How well it works:
Very well, for a proof-of-concept. 

Problem 1: The center digit is moved flawlessly by the first digit, and produces a satisfying click. But the center digit moves the last digit with only about 50% reliability. Playing with the gear timing helps, but the main problem is that Limbaugh's mechanism moves the 8t gear only about 5 teeth per rotation of the 40t, which is barely more than half way around. Modifying the gearing so that the cam was rotated further (relying less on the rubber band) would help. 

Problem 2: The timing required by problem 1 is such that when the mechanism is trying to move the 3rd digit it must overcome the full resistance of both rubber bands simultaneously. Solving problem 1 might enable the timing to be adjusted such that the return stroke of one cam assists in extending the next cam in the sequence.

-Brian Alano
http://www.kiva.net/~alano/brian