It's not a fool's errand. You are underestimating how few games most of Steam user's playtime is in. Getting proper support for ARM to make out the most performance on the most popular titles is a reasonable thing to fund. Valve can still use FEX for addressing the long tail of games, but it will have disadvantages to a proper ARM port.
But is the disadvantage worth the relatively high overhead of specifically adding arm support? I doubt that. It is better game devs focus on what they're better at - x86 - while valve and open source devs focus on what they're better at, than trying to split funds across competing solutions to the problem.
But why would Valve do that, Steam is a game market place, that happens to provide a really powerful comparability layer to allow you to run many windows games on not windows. It’s not a platform in any meaningful sense. The Steam deck is a platform, and the Steam frame, and if they can get existing games running on them, without involving the original devs what’s the problem? Dev get a new market to sell their games into, Stream gets a new market to extend their store front onto, how is that not a clear win-win?
Also Valve does fund plenty of games, such as all of the first party games you might have heard of, like Half Life, and its long tail of sequels and spin offs.