In b/67387379 it was pointed out that the call to DiscardFramebufferEXT in WebGL's DrawingBuffer is being made from a sub-optimal place. It should really be being called right after the BlitFramebuffer call which resolves the multisampled renderbuffer.

This might be a reason that some GPU architectures (Intel?) aren't seeing the performance benefits of this addition.

Note that the multisampled renderbuffer will have to be marked dirty at that point, so that an implicit clear will have to be done before it can be drawn to again.