I guess my point was just that this work seems to be presented as a significant improvement on the state of the art in RE-matching performance. As far as I can tell it's only an improvement over RE implementations that have never been held up as the state of the art for performance. Further, its performance seems to have been measured mostly on a particularly pathological microbenchmark.
Just grabbed the source code and stared at it for a bit. It's big; 14K lines of C++ code, really well commented. It does do lazy DFA compilation with a state cache, which is nice. Couldn't find any evidence of Boyer-Moore, though. This is really a major omission if performance is an issue.
Don't really have time right now to do more benchmarking of this code. There's some benchmarks there already, but I haven't figured out how to play with them yet. More if and when I get some time.