thats 2.67% of those who use steam. I'd say for every 1 who use steam, there is 2 who isn't
Its a percentage, not a number of users. Are you also suggesting that there's huge proportion of gamers out there "hardcore" enough to invest in two GPUs that don't have Steam installed? PC gaming entusiasts that don't have one single title from either Empire: Total War, any Half Life title, L4D1 or 2, TF2, Counter Strike + Source, COD:MW2, FEAR 2, Football Manager 2009/10, Dawn of War 2, Audiosurf, Zeno Clash and Killing Floor? Really, as if they own any one of those titles, then they have Steam, and even if they somehow don't own one of ths hugely popular titles despite investing huge sums of money into the platform, you think they'll have passed up every last one of those insane deals Steam's been offering?
Basically, if you're a gamer that has a dual GPU setup, its
incredibly likely that you will have Steam installed, so if anything, the actual percentage of PC gamers with dual GPU rigs is almost certainly
less than that pitiful 2.67% figure. Its a tiny, tiny niche (thank God), and just not really worth discussing in the grand scheme of things.
By any qualitative or quantitative measure, the statistics generated by the Steam surveys are the most accurate out there for the PC gaming market. You can disagree with it, but they are pretty much taken as gospel by people in the industry. They have a VERY large userbase and a significantly high penetration that they are MORE accurate then pretty much any other survey in any other industry. Nelson's wet dream would be to get as accurate data about TV viewers as Steam generates about gamers.
An excellent point. All those government polls that are discussed, appreciated and respected by some highly intelligent people usually only have a sample size of 10k or less (at least in the UK) and yet represent an even larger population. Having incredibly accurate data from a sample size of nearly 30 million active PC gamers is going to generate incredibly accurate data. There's really no feasible way of generating data that even approaches the same level of accuracy without investing a ludicrously large and unprecedented (for the field) amount of time money. Only a national census would come close and that's a system prone to human error.