I would have thought that it would be clear cut whether memory allocation is faster in managed code than in native code - but there seems to be some controversy. Perhaps memory management under a Virtual Machine be faster because of no context-swapping calls to the OS, but then I see that the VM would itself need to make periodic calls to the OS for more memory, and it has the management overheads the memory itself rather than the OS.
Rather than making unsubstantiated assertions like I have, please provide links to references that support your position.