Got the barest core of the parrot interpreter done and built, but it's enough to do simple things and give simple benchmark results. It's not even that slow. For a simple benchmark (count from 1 to 100 Million) it takes about 13 seconds on a snappy (700MHz EV6) alpha, and clocks in around 23M ops/sec. My dodgy Celeron takes 57 seconds for the same benchmark, so it's not nearly as good per clock cycle, but still... not bad.
Real source will be released to be laughed at as soon as we've got the skeleton of a working assembler. I'm more or less out of touch for the upcoming week, so probably sometime around the beginning of september.