Yeah, so I finally caught up with Erin's changes to the C++ version. It's still not quite usable though - a few bugs remain to be found, not to mention the fact patches of functionality are missing or broken. I'll be putting up a preview version soonish, though.
I also spent 10 minutes cutting out one of the more flagrant ways the engine creates too many objects (instead of re-using objects internally).
I measure the average benchmark time (using the newly created benchmark program) has decreased from 1537ms to 1378ms. That's 10% faster (as a rough estimate). Not too shabby for just one quick change. I've attached the benchmark reports.
Performance of the engine doesn't really excite me though (this is FLASH we're talking about). But it seems the done thing in dev blogs, which this is rapidly turning into.