CR2032 node and power consumption

It looks like I was way too optimistic. The CR2032 node on the balcony worked in the morning, but the battery voltage dropped from 3.07V to 2.90V, so I stopped the experiment. No need to ruin the battery when there's something not working as expected.
The node was set to send both temperature and ID packets every 8 seconds, which is definitely too frequently, but I was still surprised to see the battery getting depleted so fast. The situation called for some action. I spent most of the evening trying to figure out how to measure how long the node stays up, doing the measurements and sending the packet to central node. In the end, I did the following:
 - node was powered from 3.3V through 56Ohm resistor. That means that the voltage drops by some 700/800mV every time the node wakes up.
 - the voltage is sampled by LM339 comparator, and compared with a reference voltage
 - output of the comparator is captured using logic analyzer (Saleae clone)
This sort of works, and shows that the time for what the node is up is not consistent, ranging from a few miliseconds to 80ms (I even saw 330ms once). However, I can safely say that the time is some 50ms on average. Assuming the node taking 20mA when running, the node consumes:
50ms * 20mA = 1000uAs = 1mC (microampersecond, milicoulomb)
for one active packet. The standby power consumption is let's say 25uA (Arduino, RF24, sensor etc.), and node stays in standby for 8 seconds:
8s * 25uA = 200uAs = 0.2mC
I have to add the power consumption of the DS1820 when performing the temperature conversion. That takes 750ms and the power consumption according to the datasheet is up to 1.5mA:
750ms * 1.5mA = 1125uAs ~ 1.1mC
All together, one 9s period (8s standby, 1s temperature measurement while node is in standby, RF24 communication) consumes ~2.5mC. There are 400 periods in one hour, that's 1C of charge.
The CR2032 should be able to supply 200mAh, that's 0.2A * 3600s = 720C of charge. In other words, even with 8s between the measurements and RF24 communication, I should be able to run for 720 hours, or 30 days on one CR2032!
The only explanation I have for this is that I am simply facing two things I did not take into account:
 - The CR2032 (according to GP datasheet) drops the voltage very quickly at the very beginning of the discharge curve. The battery was brand new when I put it into the node.
 - The voltage drops a little for lower temperatures. I measured the initial voltage on my desk at room temperature, then moved the node to the balcony with temperature around freezing point.
I've changed the constants, so the node now reports temperature every 5 minutes, and sends identification every half an hour. The node now sits outside on the window ledge, showing 2.97V and -3 degrees of Celsius. Let's check it again in the morning!