The most ambitious attempt to gather into one place and analyze all known data concerning European manuscript and printed book production remains Uwe Neddermeyer's two-volume work from 1998,
Von der Handschrift zum gedruckten Buch. Neddermeyer's work was criticized both for the reliability of the data and the methods used for analyzing it, but I've always thought that, whatever its shortcomings, it was a project worth undertaking, and that the author managed to accomplish a surprising amount based on data that was broadly dispersed and not yet available in any electronic format.
So to consider the question of what the distribution of known print runs in the fifteenth century looks like ("known" here including everything from firm historical records to modern scholarly hunches), there is still no better place to turn. Neddermeyer calculates arithmetic means of a few different kinds, but I wanted to see the actual distribution.
Here's the distribution for 1450-1479, which is heavily influenced by the known edition sizes from the print shop of Sweynheym and Pannartz in Subiaco and Rome (36 of 86 editions), nearly all of which are recorded as 275 copies.
Fig. 1: Distribution of known print runs before 1480
In other words, if not for Sweynheym and Pannartz, we would see the greatest number of editions below 250 copies, followed by those betwen 250 and 500, and with another group at 750-1000. The dominance of Sweynheym and Pannartz, coupled with the small number of known edition sizes from any other source, makes the evidence difficult to interpret with much confidence.
Fig. 2: Distribution of known print runs, 1480-1500
After 1480, the data sources are somewhat more varied. Interestingly, there's still some sign of a bimodal distribution with one peak in the range of 250-500 and another in the range of 750-1000. (A few broadsides with print runs in the range of 3,000-20,000 copies are not included here, but would not amount to a bin of any notable height.) Having a sense of what the distribution possibly looked like is often more useful than establishing a precise figure for an average print run, particularly if the average print run was not actually all that common.
How representative is Neddermeyer's sample? His table of known fifteenth-century print runs (2:753-62) also includes the number of surviving copies where a specific edition could be identified, and the number of surviving copies suggests we should treat this all with some caution. For the 152 items in his sample with a known number of surviving copies, the average number of copies is 40 - or almost three times the average of 14.4 copies for all known incunables. As averages are especially in this case not nearly as useful as distributions, let's take a quick look at how the number of surviving copies in Neddermeyer's sample is distributed. The initial graph is messy:
Fig. 3: Distribution of surviving copies from Neddermeyer
Even here, the 0- and 1-copy editions are the most frequent. We can make the data easier to visualize with a histogram with bins that are ten units wide.
Fig. 4: Distribution of surviving copies from Neddermeyer (bin size = 10)
And so the customary distribution of incunable survival emerges again. The predominant number of copies is 1-10; the green bar to the right comprises the editions with zero known copies.