Clock #1 has been in our dining room since I moved it upstairs since March 11. It ran for several months, and then in mid-April started to become unreliable. I suspected weather factors were the cause. After running from mid-April until mid-May, it stopped again, this time apparently for good. After letting it sit for about 10 days, I went back to investigate the issue.
The overall friction appears to be higher in the spring/summer (though I don't know for certain), although it's unclear exactly if this is localized to a specific mesh in the train..
It seems that the problem was that a few escape teeth were too long by a very small amount (8/1000"). When the friction is lower in the train, the pendulum amplitude is high enough that this doesn't matter. But when the amplitude drops, this becomes a problem.
To debug the problem, I stuck a post-it note to the back of the frame behind the pendulum. I marked on this paper the precise pendulum location when each escape tooth released. The marks were about 13" from the pendulum pivot, and indicate that the typical distance between entry and exit release was 3/16", or about 0.8 degrees peak-to-peak. So if the pendulum swings less than that, the clock is likely to stop. I found that one tooth that seemed to cause stoppage consistently corresponded to a mark 1/16" farther out from the rest. This means that an additional 0.15 degrees was necessary to escape that tooth. Since the anchor has a length of 1 3/4", this translates to an escape tooth of about (1/16)*(7/4)/13 = 8/1000" longer than the rest. That's a very small amount, but easily corrected with a file. Once I corrected that, I checked each other tooth as well, adjusting them so they were all within the 0.4 degree peak-to-peak release margin.
It seems unlikely that expansion or contraction of the escapement itself (due to weather) is the cause of the clock's malaise, especially because that particular tooth was already marked as being problematic in the past! Hopefully, my guess about balance amplitude is correct, because it seems to also explain the other issues about the escape wheel sitting at the front/back of the clock too. This slightly shifts the escape wheel up/down by a very small amount and seems to change the effective length of the escape wheel teeth by a few thousandths of an inch.