Release source code to antiquated games (gems) like half life 1, doing so will allow the community to keep these games alive for as long as users are around. Data content is still required so, they still can make revenue (and maybe even more since customers know they can play the game on any device they own)
Valve can push things to happen more than the average linux user can. They could push for documentation and opening of device drivers more so than the end user could ever do. They could finance open source developers like nouveau .
Contribute to KDevelop / QtCreator to make the platform more attractive for developers used to Visual Studio. While I personally can't stand VS, in the research lab here it is basically the only reason most are using Windows.
While QtCreator is very solid, it's code analysis is still somewhat basic.
KDevelop on the other hand has some very nice features here and there, but it unfortunately still is, in some circumstances, too crashy or unreliable to really be recommendable for productive use.
I agree with pickle, open sourcing their Goldsrc back catalog would be one huge thing Valve could do. They are hardly expected to open source Steam itself since that is a DRM platform and they are still actively using Source for current and upcoming games. But Goldsrc is older and so they won't be revealing any precious secrets by open sourcing that. The fact that it is based on the Quake 1 engine shouldn't be a problem since I am sure that id Software would gladly allow them to do it since they themselves have already released the Quake 1(and Wolfenstein 3D, Doom 1/2, Q2, Q3A, Team Arena, RTCW, RTCW:ET and Doom 3 for that matter) source code.
They could finance open source developers like nouveau .
That's somewhat of a misconception... I put Valve in contact with Nouveau [Mupuf] months ago, but that they don't really need/want money. Their developers are currently still at university (and don't want to quit) or are happily employed elsewhere, so first they magically need to find some qualified developers. It's also somewhat also the case for Radeon; I tried to get Marek to talk to Valve, but alas he wants to go to university for another year.
I think the greatest front Valve can help with is getting their games to run as fast as their Windows counterparts can, as stated in the Steamed Penguin post. This would mean a lot of things would need to fall in line. Wayland should be targeted at, they said Ubuntu was their target, but I'm not sure about the situation regarding commits from external entities like valve. Unity3D is shown time and time again that it is fundamentally bad for games.
The second most obvious point would be to get valve to lean hard on both amd and nvidia to focus most of their attention on the open source drivers and get the latest graphics engine goodies running on them. If valve needs to hire a full timer to help complete this work, they should. Valve's best bet is to get their games running on Linux as good as windows, if not better. This will lure a bigger user base.
Unfortunately, valve can't do everything, but I think what's needed is integration. Just make the game work, and bring everything along with it. Like audio, controller input. Heck, even go crazy and make a steam desktop environment to boot into at login. Why stop there? Go for the steam box now. Make that distro and optimize all the things!
Just by doing their job conscientiously and solving/fixing the problems they come up they'll benefit everyone. I'm sure there are a plethora of all kinds of bugs and problems with the drivers and the system. If they can fix some of those and document the rest they'll do a lot more than any conscious effort to improve something in Linux.
Another thing they could do that they didn't do when they came to Mac OSX, would be to port the Source SDK, Hammer and the associated tools(Such as map compilers, importers/exporters, facial animation/lip synching tools, etc) to Linux. Also just in general encouraging others dev studios to support Linux. The big question is how they could do that? Trying to get as many games as possible to launch when Steam arrives is one obvious thing. I guess the #1 priority in that context is getting games that already have Linux versions and are on Steam to launch. Hopefully they will be extending SteamPlay to cover Linux for games with Linux ports and the neccesary rights in place. What other steps could Valve take to encourage others devs to support Linux?
They have done a lot already by generating interest in and thus creating potential new users for GNU Linux. I think the best thing they could do is advertise on the steam client the same way they advertised for the osx client. Many people have no idea that linux is even an option. Any exposure by such a respected entity will do a lot of good.
Since I starting using linux (with Fedora Core 5), the majority of people i showed it to hadnt heard about it. Today at work someone asked me about it (as the steam linux news was tech news front page on yahoo.com this afternoon). After Ubuntu 8.04 it has been a very good alternative for casual users and most of the non gamers i have shown it to still use it.
So in my experience there are two problems. Lack of AAA games (hopefully solved soon) and exposure. Valve has the resources to impact both problems significantly.