tcmalloc
I'm considering using the Google tcmalloc library for handling memory. Our libevent core library is really memory hungry, so maybe tcmalloc would help here... well, it is that way because we do a lot of buffering for reducing the CPU load, so it is not really a libevent fault but the way we use it.
I've looking for experiences with this library, and it seems the Squid guys have dome some work with it. They were having some memory fragmentation and performance problems, so they replaced malloc with tcmalloc and they obtained pretty amazing results.
In fact, the whole Google Performance tools look very interesting. They have a full suite of utilities for profiling C++ apps by simply LD_PRELOADing their libraries, and then you get data that can be processed and generate nice CPU or memory usage graphs. They even have a library that, once it is preloaded, starts a web server on your app and you can profile it remotely...