Kernel code shouldn't need a 'good and fast' allocator though, since preferably it shouldn't heap-allocate in the first place, or at least not frequent enough that allocator performance matters.