Using async methods of TlsStream causes ThreadPool starvation

[jivkopiskov]

The current implementation of the Tls.Stream async API (BeginRead, ReadAsync, etc) just schedules a task and calls the blocking non-async Read.
This causes most of the thread pool threads to be blocked with a stack similar to this one:

  System.Net.Sockets.il!Interop+Winsock.recv(class System.Net.Sockets.SafeSocketHandle,unsigned int8*,int32,value class System.Net.Sockets.SocketFlags)
  System.Net.Sockets.il!System.Net.Sockets.SocketPal.Receive(class System.Net.Sockets.SafeSocketHandle,value class System.Span`1<unsigned int8>,value class System.Net.Sockets.SocketFlags,int32&)
  System.Net.Sockets.il!System.Net.Sockets.Socket.Receive(unsigned int8[],int32,int32,value class System.Net.Sockets.SocketFlags,value class System.Net.Sockets.SocketError&)
  System.Net.Sockets.il!System.Net.Sockets.NetworkStream.Read(unsigned int8[],int32,int32)
  BouncyCastle.Crypto!Org.BouncyCastle.Tls.RecordStream+Record.FillTo(class System.IO.Stream,int32)
  BouncyCastle.Crypto!Org.BouncyCastle.Tls.RecordStream+Record.ReadHeader(class System.IO.Stream)
  BouncyCastle.Crypto!Org.BouncyCastle.Tls.RecordStream.ReadRecord()
  BouncyCastle.Crypto!Org.BouncyCastle.Tls.TlsProtocol.SafeReadRecord()
  BouncyCastle.Crypto!Org.BouncyCastle.Tls.TlsProtocol.ReadApplicationData(unsigned int8[],int32,int32)
  BouncyCastle.Crypto!Org.BouncyCastle.Tls.TlsStream.Read(unsigned int8[],int32,int32)
  System.Private.CoreLib.il!System.IO.Stream+<>c.<BeginReadInternal>b__38_0(class System.Object)
  System.Private.CoreLib.il!System.Threading.Tasks.Task`1[System.Int32].InnerInvoke()
  System.Private.CoreLib.il!System.Threading.ExecutionContext.RunFromThreadPoolDispatchLoop(class System.Threading.Thread,class System.Threading.ExecutionContext,class System.Threading.ContextCallback,class System.Object)
  System.Private.CoreLib.il!System.Threading.Tasks.Task.ExecuteWithThreadLocal(class System.Threading.Tasks.Task&,class System.Threading.Thread)
  System.Private.CoreLib.il!System.Threading.ThreadPoolWorkQueue.Dispatch()
  System.Private.CoreLib.il!System.Threading.PortableThreadPool+WorkerThread.WorkerThreadStart()

Is there a way to use the TlsStream in a non-blocking way? As far as I understand, BC would need to add proper implementation of the async apis for the TlsStream, TlsProtocol and RecordStream, which will use the Async APIs of the provided inner stream.