How do you think this compares to Perl::Mason? How are you avoiding the same pitfalls that Mason has? complicated failure modes, pretty much impossible to improve timing and create on demand resource prefetching without increasing complexity greatly?