It's easy to come up with cases where you can do better than a fairly reasonable programmer: think of a loop that allocates at the top and frees at the bottom with enough stuff happening in between that the allocator doesn't just hand you back the same memory from a cache. Depending on the runtime, it might be better to just bump allocate and garbage collect all that memory at the end. You can find the benchmarks that'll show it, too. In reality, with actual applications garbage collectors have "good enough" performance with many benefits such as correctness and ease-of-use, but manual memory management done by a skilled programmer will outperform it.