Actually.. I've been doing some perf measurements in the past few days on b2 engine and I was rather surprised where the time goes (not the memory). For example, currently, the most expensive operation is in switching module contexts. Specifically here <https://github.com/boostorg/build/blob/develop/src/engine/function.c#L4169>. There are many reasons for why that's slow (some obvious from just looking at those few lines). But the key problem in the current implementation is that other than the optimization work Steven did we haven't done design changes for optimizations. But that's something I was about to get into in an imminent future email :-)