An online article discussed a subtle behavior on OS X, NetBSD, and OpenBSD delivering the SIGPROF signal to multithreaded programs: http://research.swtch.com/macpprof. In a nutshell, the current implementations in those systems cannot do it right to deliver the signal to a running thread, which most other systems would do.

This behavioral difference affects profilers that rely on SIGPROF to avoid sampling bias. For example, we can see a short description here: https://code.google.com/p/lightweight-java-profiler/issues/detail?id=2.