A frontal attack to Windows is clearly impossible, other strategies, synergically implemented, could work.
Amongst possible strategies:
1) A necessary step should be in any case to help making OpenGL more attractive than DirectX for devs, providing complete and updated libraries, dev tools, environments and kits, and possibly easy porting tools too
2) SteamBox
3) Possible synergies with Android, to make SteamBoxes and PCs running SteamOS an attractive higher-end evolution of current Android devices
4) Making Steam a true cross-platform gaming environment for as many architectures as possibe and viable
5) Proper marketing
6) Deceiving MS, not only never attacking it frontally, but also doing everything possible to make it perceive the threat as late as possible.