I think this is a bug in the handling files that have changed on disk and needing rescan. There’s a constant stream of failed requests from A triggering constant rescans on B causing constant updates to A which retries and requests from B… Maybe the rescan never happens.
There was a fix for something similar in 1.6.0. You might try that.
One side requesting a constantly changing file means it gets rescanned as often as it’s requested, i.e. as fast as the ping-pong goes (50ms is quite impressive). Not sure whether there’s anything for Syncthing to do here, syncing a moving target won’t work anyway.