1) Linux/PPC's gcc toolchain has supported VMX (AltiVec) for a very long time, but there aren't many Linux/PPC developers to begin with and they typically use older Macs so there hasn't been much development. Some codecs may use VMX for FFT and whatnot.
VMX is actually pretty easy to optimize for. You can even do it in C as opposed to assembly. You can also use a compiler known as VAST to auto-vectorize your C or FORTRAN source for VMX and/or SMP.
2) Not yet. Somebody on this forum speculated that dual G4 cards for Macs would work, I *highly* doubt this is true.