I've just pushed a branch to my fd.o tree that brings back the standalone compiler. I've also made some small enhancements to it. Why? Thanks for asking.
One thing that game developers frequently complain about (and they're damn right!) is the massive amount of variability among GLSL compilers. Compiler A accepts a shader, but compiler B generates errors. Then the real fun begins... which compiler is right? Dig through the spec, report bugs to the vendors, wait for the vendors to finger point... it's pretty dire.
Mesa compiler has a reputation of sticking to the letter of the spec. This has caused some ruffled feathers with game developers and with some folks in the Mesa community. In cases where our behavior disagrees with all other shipping implementations, I have submitted numerous spec bugs. If nobody does what the spec says, you change the spec.
This isn't to say our conformance is perfect or that we don't have any bugs. Reality is quite the contrary. However, we are really picky about stuff that other people aren't quite so picky about. When we find deviations from the behavior of other implementations, one way or another, we sort it out.
Sometimes that means changing our behavior (and adding piglit tests).
Sometimes that means changing our behavior (and getting the spec changed).
Sometimes that means implementing a work-around for specific apps (that is only enabled for those apps!).
Sometimes that means not changing anything (and taking a hard line that someone else needs to fix their code).
The combination of our ability to build our compiler on many common platforms and our spec pedanticism puts Mesa in a fairly interesting position. It means that developers could use our compiler, without the baggage of the rest of a driver, as the thrid-party to settle disputes. It can be the "if it compiles on here, it had better compile anywhere" oracle.
Even if it fails at that, we emit a lot of warnings. Sometimes we emit too many warnings. A standalone compiler gives a good start for "lint" for GLSL. There are already a couple places that we give portability warnings (things we accept that some compilers are known to incorrectly reject), so there's already some value there.