There shouldn't be any way for the decoder to decode something and the UI not be told, so the problem is not there.

What might be happening is that the pipeline is forked, where the camera itself it at the root and one branch goes off to display on the screen, and the other is piped through the libzbar and libdmtx decoders (implemented as gstreamer elements) one after the other.

Now the framerate is higher than the decoder rate, so some of the frames displayed on the display branch will simply be dropped by the decoder branch until the decoders have finished processing their current frame.

So to answer that question, no, the frame shown is not necessarily the one decoded, but there shouldn't be much lag (though there is more lag on dmtx than on QR codes as the libdmtx decoder is slower).

Speaking about decoding from blurry images, the QR code decoder really is rather good at this, better than the 1D code decoder certainly.