For many small stream writes, RecyclableMemoryStream is slower than other implementations

For a project which implements its own buffered memory stream I added the int Read(Span<byte> buffer) and void Write(ReadOnlySpan<byte> buffer) signatures and compared the performance of MemoryStream, ArraySegmentStream (custom based on ArrayPool) and RecyclableMemoryStream.

To my surprise, RecyclableMemoryStream is always outperformed by MemoryStream and ArraySegmentStream by a factor of two, as if the Write(ReadOnlyMemory<byte>) is not properly implemented or disabled in RecyclableMemoryStream?

see benchmark results here: https://github.com/OPCFoundation/UA-.NETStandard/pull/2556

see results of BinaryEncoderRecyclableMemoryStream vs BinaryEncoderArraySegmentStream and BinaryEncoderMemoryStream.

Any idea what could be wrong in the benchmark or why RecyclableMemoryStream.Write is always slower?

Start with the benchmark results in OPCFoundation/UA-.NETStandard#2556, comparing BinaryEncoderRecyclableMemoryStream with BinaryEncoderArraySegmentStream and BinaryEncoderMemoryStream. Trace the Span-based Read and ReadOnlySpan-based Write paths in RecyclableMemoryStream and verify whether the benchmark exercises them. Done means identifying whether the benchmark is flawed or documenting the cause of the performance difference.