Media SDK tutorial provides with samples for using D3D acceleration for decoding and for decoding several streams simultaneously via OpenMP.
But is it possible to use D3D performance improvement if I have to decode several H.264 streams simultaneously?
You can run many decode sessions simultaneously without issues. The total number of sessions you can run depends on the amount of available memory and DirectX resources.
The underlying hardware will share the available resources evenly between the concurrent decoding (or encode, VPP) sessions. Naturally, since the hardware resources are shared, the individual sessions will run at decreased speed compared to the case where only one workload is executed.
In the tutorial, OpenMP was used to illustrate concurrency in an easy way, but you can opt to use any other available concurrency methods fitting your needs.