[boost] [acc] FFT-aCC results surprisingly good?

A new regression engine for HP-UX acc has just entered (FFT-aCC), and the success rate is surprisingly high, given that: 1. The compiler has made it very late to the Boost 1.33 scene --some complex new libs, like Boost.Parameter didn't have previous exposure to the compiler AFAIK and yet they work flawlessly. 2. Previous versions of aCC were notoriously buggy --for instance, Dave&Aleksey's MPL book lists HP aCC 03.55 as a lost case. Now,