I believe in the context of WebRTC and video chat generally this is less relevant, as many hardware implementations are limited to the "play a video" use case, and while in theory they may have enough silicon to decode 4 small videos of other participants, while encoding another of the user, all at the same time they don't often in practice support that use case.

When it was discussed in one of the WebRTC meetings the only video chat application that anyone could name that used hardware acceleration was Apple's Facetime, if I recall correctly.