Results on one benchmark:
10% saved on CPU time,
10% saved on actual run time,
20% saved on RAM max usage,
20% saved on disk writes
The benchmark really is on the internals of reaction, so it doesn't take into account real life scenarios where reaction has more complex regexes, and writes more complex data in DB.
So this is kind of the theorical maximum performance boost this change can make.
Still, that's a lot 馃弳