Bluh. I particularly dislike estimates of this kind which are obviously much, much more exact than the actual data permits.
From the given data, anything between 2 million and 10 million could be perfectly consistent. The former if the re-buy rate is higher and the average number of players per DM is lower (e.g., if it was typical for players to also become DMs down the line.) The latter if the re-buy rate is low and most DMs had more than four distinct players.
Specifying down to individual people is silly. Just call it 4 million and be done with it. You're making assumptions which shift these values by hundreds of thousands, specifying down to thousands, hundreds, tens, and ones is false precision.