I could totally understand why the Nest team wouldn't view it as a compelling product requirement. You need the cloud anyway because people regularly control the thermostat outside their home (e.g. to pre-warm/pre-cool when they're heading home). It's all the same infrastructure that's leveraged to run products beyond the thermostat (doorbell, security) that also have a compelling remote use-case. The isolated LAN approach could still require a cloud component for managing access. The C&C UX could become quite confusing trying to support both isolated LAN & cloud. If the cloud part fails you're no worse off than if you had a non-smart thermometer/doorbell with a good UX.
Now you could obviously have remote abilities without the cloud but the UX story becomes even trickier because you need to figure out how to rendezvous & punch through a wide variety of firewall configurations. These are all things technical users could solve given the appropriate settings exposed (eg. dynamic DNS) but it's totally unnecessary for the less technically minded users Nest is targeting (or even technical-minded users who don't want to care about managing every single smart device on their network).
I'm also unclear why you are claiming the only gains are in maintenance. Security is a real end-user benefit & having devices not allow arbitrary incoming network connections is a security choice. Similarly, arbitrating everything through the web results in a simpler, more reliable, enrolment process & makes it easy to add customer features like family control more easily, securely & reliably. These are all customer-facing advantages.
2 hours out of a year is still 3, almost 4, 9s of uptime. That's probably better than your home network/ISP.
I get it. You personally would prefer they traded off for supporting direct connection to your phone. However, that doesn't invalidate that there are legitimate engineering reasons someone might choose other tradeoffs.