I took a gamers perspective to sorting this, anything in the signal line that delays the video input is bad. IMO it's video input lag. The 3700 chokes on high bandwidth video data. These things are designed by old school AV engineers who don't have the same critical specification parameters that gamers are looking for.
I noticed that lip sync was improved by switching my ATV to 4:2:0 or the reciever output to "standard" instead of enhanced or 4k/8k.
Then I noticed that the receivers video processing is set to "auto"...well if game mode reduces video input lag...then entirely bypassing any video processing might work? So I set the reciever to bypass.
The delay has gone to all intents and purposes - in a blind test my wife can notice the lag when recievers video processing is set to auto and can't notice it when video processing is bypassed. That is for all apps on the ATV.
The receivers input lag is really, really bad in anything other than game or bypass mode - and even then only the latter seems to cure the problem entirely.
If these engineers want to add heavy processing at every stage of the signal pathway, (this goes for audio as well) whether it's in the reciever or TV - they need to find a reliable way to make the devices communicate to each other what those processing delays are - for example - if I enable truemotion - the video lag gets a little worse again. This means the TV would have to do a some kind o handshake with the reciever everytime you changed picture mode, app or input...which I doubt it does and it would be non-trivial to implement because of all the lag when switching apps or inputs, to get the delay measurement and route that back to the reciever. That's probably 250ms added to opening each app or change of video format. This is a case of features rapidly multiplying without any FMEA being done.
Further, even if the TV is doing Dynamic Lip Sync, I'm definitely sure that the crappy processors on board the Denon can't do it.
Moral of the story, processing across multiple devices is nasty and shouldn't be done, 24/25fps video needs to fucking die so we don't have all this truemotion stupidity and general motion handling needs to be hardware focussed not processing focussed.
The other way to resolve this would be for Sony to get back into the AVR game and make something that really works with current TVs.