Of all experimental public good games that have been conducted in laboratory & field settings, what is the average percentage of subjects that can be classified as free-riders (as opposed to (strong) cooperators & conditional cooperators)?
So you are looking for a meta paper that aggregates results from other papers. What you mean by "all experimental public good games" is unclear because the setup of most public good games is different. The multiplicator of the public good, the number of rounds, the level of allowed commucination, whether there is punishment, etc., these frequently vary and have a very big impact on the course of play.
Another reason why such a meta paper is unlikely to exist is because the ratio of cooperators is context dependent. How the possible actions are labeled: 'Cooperation' and 'Defection' or 'Give money' and 'Save money' has a strong effect on behavior. So even seemingly similar, 'basic' games could be very different.
Your best shot may be trying to find a famous experiment and seeing if there have been any replications. You could then aggregate the data and calculate the desired ratio.