The short answer is **memory fragmentation and allocation granularity**.

While a single large buffer seems simpler (one allocation, one