The main thrust of the neoreactionary argument, as I see it — or, at least the part that interests me — is that the cost of a state failing is so catastrophically large that it's really worth looking critically at how various forms of state-level organisation have emerged, risen, and fallen in the past. After all, we have actual data now on how monarchies, republics, dictatorships, and so on have weathered war, environmental disaster, economic instability, domestic instability, and other disasters — or not, as the case may be. Why shouldn't we use that to inform our decisions about how we build resilience into our social structures? Buckminster Fuller would be proud.
What I question, though, is one of the fundamental assumptions: is it necessarily all that costly for a state to fail? I'm pretty convinced the answer to this question is "no, depending on how the state is set up to handle failure." But in order to explain why, first let's talk about mail servers.
Dan Bernstein, aka djb, is the author of several highly regarded implementations of various useful things, including a DNS server, djbdns, and a mail server, qmail. The services djb writes are lightweight, secure, and not as widely used as they ought to be. That last is due to a principled design decision on djb's part that unfortunately conflicts with a lot of people's intuitions about how software ought to work: crash-only design. Crash-only design is a brute-force-and-ignorance way to satisfy Tony Hoare's dictum that the most important property of a program is that it accomplish the intentions of its user: if a process hangs, or otherwise enters an inconsistent state, it dies (or is killed by a watchdog process) and respawns in some previous known-good state, rather than trying to flail its way from somewhere in failure-space back to working order. Anything exceptional is regarded as antithetical to accomplishing the intentions of the user, as the user's intentions can only be accomplished if the service is working normally. "Time to working order" is therefore one of the variables in a process-failure scenario that you want to minimise. If you don't know how long it might take a service to recover from exceptional circumstances, it's hard to plan around that: how many inbound requests do you queue? what do other services that need it do in the meantime? But with crash-only software, "time to working order" is however long the process takes to respawn. You want it not to crash at all, but if it loses the plot, you want it to give up and restart quickly to avoid downtime.
"Brute force and ignorance" isn't an insult there, incidentally. No matter how stupid an approach sounds, if it works, it isn't stupid. There have even been serious, well-reasoned proposals for building entire crash-only systems, composed from components which understand "on" and "off" and are designed to flip from one to the other as quickly as possible.
I bring all this up because Belgium effectively has a crash-only cabinet.
The Kingdom of Belgium is a parliamentary constitutional monarchy, politically well to the left of the US but somewhat to the right of Scandinavia. Like the United Kingdom and the Netherlands, our monarch's duties are mostly ceremonial, but unlike either of these, our monarch has also had to perform non-ceremonial duties in recent memory.
We've talked about Duverger's law here before, with respect to its failure case that the US experiences: in a first-past-the-post electoral system, "successful" voting eventually comes to mean "voting for one of two major-party candidates," and eventually the legislature settles into 50-50 deadlock. Belgium gets the other one. Seats in parliament are allocated proportional to the number of votes that various parties receive, which leads to a proliferation of smaller parties. The members of Parliament then have to figure out how to form a coalition of at least 50% of the seats, with the extra hitch that any seats filled by a member of a party with a cordon sanitaire around it still counts as a seat, but nobody will accept that member into a coalition. If they can't reach that 50% threshold — including if a party pulls out of a coalition later — then we don't have a cabinet and everything deadlocks.
Except we also have a watchdog process, and its name is "the monarch." If Parliament can't form a coalition, one of the royal powers that's effectively ceremonial under normal circumstances suddenly becomes really, really important. After every federal election, the monarch normally appoints a formateur to manage the process of appointing cabinet ministers. Typically the formateur is also the person expected to become prime minister, which is why this is normally a formality. But if a coalition collapses, the monarch can nominate a new formateur to establish a new cabinet, sometimes repeatedly. The cabinet was down for 589 days in 2010-2011, and it took eight tries to reboot it.
The really interesting thing about this is that not a lot of people seemed to mind all that much. A lot of politicians had soundbites about how embarrassing it was, and given that between 2007 and 2011 we had no cabinet for a total of over two years your odds of reading a "Belgium still doesn't have a government" article were pretty good if it was a slow news day, but that was about it. The postal service still operated, the military (and federal police, which we have and the US doesn't) still got paid, national parks were still open — none of the infrastructural hostage-taking that's starting to become business as usual in US budget deadlocks happened here. Beer and chocolate did not instantaneously morph into warlords and AK-47s simply because the cabinet process failed. If the executive is the State, then by that definition, Belgium has failed repeatedly within the lifetime of the average third grader.
A pedant, of course, would point out that since the monarch is the official head of state, no constitutional monarchy can fail unless the monarchy also fails. I'll accept that, and it's certainly not lost on me that I started out talking about neoreaction and am currently tolling the advantages of monarch-as-watchdog-process. I'm not sure what happens if something triggers a succession crisis during a coalition collapse (let's assume the least convenient possible world, where the entire succession hierarchy gets taken out in one shot like in that one John Goodman flick). In reality, though, it's probably something rather prosaic, like $MOST_SENIOR_SURVIVING_MINISTER stepping up into the functional role of "person who appoints the formateur" and business continuing as usual.
My point is: the day-to-day business of governance is all about continuity planning. If your government has taken responsibility for roads, or mail delivery, or power generation, it is the job of a wide variety of different people to make sure that the roads stay driveable, the mail gets delivered, or the lights stay on. More importantly, though, the processes they carry out are functionally, and should be operationally, separate from other processes like "what a cabinet does." Enforcing operational separation is a win for stability precisely because it disincentivises hostage-taking, but also because it facilitates crash-only design: if a service is becoming unstable, you bounce the damn thing, and you don't endanger the stability of the rest of the system.
Where I part from the neoreactionaries is that I don't think you necessarily have to have a monarchy, or really any form of autocracy, as the watchdog for crash-only governance to work. But this gets into a long comparative analysis between, at the very least, the Belgian, Dutch, and British monarchies, and so I think that will be another story for another time.