benchmark on binary trees

[visitor via Digitalmars-d-learn]

ok, i have a working version (memory is nice, twice the speed as 
non parallel) ;
http://dpaste.dzfl.pl/504a652c6c47

real	0m14.427s
user	1m19.347s
sys     0m0.124s

i've got similar performances, without Allocators, using directly 
malloc and free

i had to recursively deallocate ...

though, still far from competing with the fastest, any advice ?
i guess FreeList allocator is not the better tool here ?