Quoted from DirtFlipper:

One meter counts coin drops, the other counts game starts. The delta is 'free games' (at least that's how I remember it). So one meter might read 80,000, and the other 60,000, indicating that of the 80,000 game starts, 20,000 didn't need a coin drop (i.e., they were replays). (I've got a sample Quick Draw, and that's about what the two meters read.)
Sample games were introduced to locations ~6 months ahead of production, to get additional play testing (I base this on looking at the date codes on the score motor casings between sample games and production games of the same title). Based on operator reports (they had a form to fill out; I've got a copy of one somewhere), Gottlieb could then tell how well the game was doing, how easy/hard it was to win games, if any additional bugs existed that needed to be addressed. Then adjustments could be made prior to production. Sometimes the changes were pretty significant, including artwork changes. The 50's/60's games seemed to have more changes between sample and production games than did the 70's games though.
It's a fun exercise to spot any differences between sample games and production ones. Some are fairly noticeable, others are quite subtle, and the later games usually don't have any.