At this price point, for many projects it doesn't matter; you can now use Raspberry Pi as a throwaway SOC for powering little stuff. I wouldn't, for example, put an $35 RPi 2 in charge of a bunch of sensors over my doors, but with $5 RPi it is an option.

EDIT:

And given the power, you can start offloading computation on what would otherwise be "dumb" control/data acquisition units. Hell, with that much compute you can start running serious DSP on it, which opens some additional use cases.

The only case where it matters for me is that this runs Linux, which means that I'll have an extra computer to secure against intruders. An Arduino can't do as much damage to the rest of the network if it gets hacked.

Linux also isn't a real-time OS which is often important when dealing with motors and sensors.

However, you can get very far with the built-in real-time capabilities. If your software stack can deal with occasional jitter in sensor updates (and normally you have to take that into account), you can compile a stock kernel with PREEMPT_RT and can use C to build software that has to do IO every 1ms. In POSIX. That is huge.

What would be a real-time OS? I'd be interested in hooking up a bunch of independent boards that could interact with each other, and have as quick as a response to sensors as possible... Do you have any advice as to where to start with this type of project?

Sensor latency will be measured in milliseconds in Linux. A real time OS is more necessary when you have feedback loops, inverted pendelum problems, bitbanging protocols, etc. Not being a real time OS just means that when you say "ping me in X time" it might ping you in X +- Y. Y tends to be too big for critical applications, but generally small enough that it won't matter.

Don't plug an ethernet adapter or wifi dongle into your Raspberry Pi Zero if that's your problem.

How would it communicate?

It could run RiscOS instead, with a very different attack surface?