But really, any kind of battery, even the best one, will have more self-discharge than a few nA. That makes no sense.
I think people can afford a coin cell every couple of years 
[....]
No one else noticed that the ultra low power stuff seems to break before you change the first battery anyway??
A lot of electronics is embedded => digital => gets outdated even if it still works. Even in the space of RF where everything moves a bit slower, new regulations will require parts to require to use less transmit power, bandwidth, shorter duty cycles, tighter channel integrations etc. Old protocols will be phased out, don't scale properly, etc.
That makes 'ultra low power' IMO more about not being wasteful. Another aspect of not wanting to replace batteries is the effort and irregularity to do so. Certainly its not the first thing you think about if something stops working after 8yrs..
(my 2 cents after working on battery-free stuff for a few years)
You can make any microcontroller this low power by adding a P-FET on the Vcc rail. When you want to go to sleep let go of the gate, the FET turns off and cuts away your power supply. So your power consumption is now whatever the leakage of the FET is (and that can be in sub nA) even if the lowest power mode on your MCU is a huge >10mA
To be fair this is not much different than that lowest power state on MCUs where pretty much the whole MCU goes dead and waits for a outside signal to wake up. So for this reason it is absolutely useless to look at the lowest possible numbers. You can get the same functionality using an external FET,
The interesting part for low power MCUs is how many features of the MCU continue working in the really low power sleep modes. Like timer wakeup, brownout detection etc.. Also a big factor is how fast it can wake up. It doesn't matter how low power your sleep mode is if you burn up 99% of the power during the wakeups that take too long to do something useful and return to sleep.
Off-the-shelf discrete MOSFETs that leak < 1nA are not that easy to find. Really. And you need to read the fine prints and test it - usually that degrades rapidly with temperature, for instance.
With that said, yes, unless you have an extremely low duty cycle, what will matter is how fast you can wake up and what keeps running in this low power state. For that, ESP32 in general are hopeless, as an example.
But again, even so, I still fail to see with what kind of power source a difference of say 1nA or less will matter at all. If anyone can point me to an example of such power source, I'm a client!
But really, any kind of battery, even the best one, will have more self-discharge than a few nA. That makes no sense.
Back of the envelope:
300mAh coin cell that loses 0.5%/month is 1.5mAh/(30*24h) ~ 2uA drain
But for those long-life applications you use LiSOCl2 and
https://www.fdk.com/battery/lithium_e/coin_primery/ says they do 1% a _year_, which is closer to ~350nA.
So, a few nanoamps here and there is unlikely to do much unless you're using the smallest of cells, where the self discharge current starts to be on the order of tens of nanoamps.
And you need to have saved the current state into flash or EEPROM before sleeping, and re-initialise from that.
The MSP430FRxxxx devices would beg to disagree. 
On 2 nA? Link to device data sheet please.
All you need is a power switch and it's 0nA. Flip the switch, initialize, and continue on - no need to "save the current state into flash or EEPROM", no need to sleep either, just power off.
Yeah going down very low means looking for some more specialized FETs, A lot of BJTs have even lower leakage. But most transistors in general are not specifically rated for it in the datasheet. And even when using a special low leakage transistor, something as simple as not properly cleaning the assembled PCB in clean methanol might be causing more leakage than the transistor.
But as said above A few nanoamps up or down is not really going to make a difference when most of the power sources like batteries or supercaps will likely have a higher self discharge rate than that. Besides if you use one of these super low power modes where the MCU is basically dead, you will need some extra circuitry that wakes it up, so you have to budget in the power draw of that.
When we had to make a annoyingly power hungry ESP32 run for days off a supercap we ended up cutting off its power supply and then having an external low power RTC wake it up every so often. Worked great.