A better supply of games in Linux would lead more gamers to run Linux. More gamers running Linux will make the platform more attractive to developers. This would lead to more games in Linux, which would lead to more gamers running Linux, etc. This will also lead to improvements in Linux distributions for gaming, in terms of built-in features and hardware support, making the platform more attractive to developers and gamers, and so forth.
It's just a question of where the tipping point is. This move by Valve is a big push in the right direction.
Carmack might be right that for most games, producing a Linux version would cost more than it brings in, but in the long term an investment in making Linux a viable platform could increase future profits and promote brand loyalty toward the companies that had the foresight to pioneer. I think Apple has shown us that you just can't overinvest in "coolness". And the cost-benefit problem probably doesn't hold true for less technically ambitious games, which can achieve Linux compatibility for minimal cost if their code is reasonably well structured and segregates the platform-specific bits.