OK, here is my next and final theory. People often believe that HDMI pass-through is a simple in-and-out process, but that's not right. Your AVR has to analyze incoming HDMI data to find relevant commands. My theory now is that RPi's volume change data are challenging for your AVR's processor. Probably so challenging that pass-through of video HDMI data will be dropped for the period of volume change.
However, my advice stays the same: get the extractor. Then your AVR just has to deal with audio data.