It's very easy. Australia regularly does it with millions of ballots.
Step one: you count the ballots by hand.
Phew, what an exhaustive list of instructions!
So you either have the communication between every round (which is essentially re-counting) or you physically take all the ballots to a central counting location, which introduces delays and chain of custody issue.
Also, IRV leads to more spoiled ballots. http://scorevoting.net/SPRates.html
So while there are more important reasons to prefer Score Voting and Approval Voting (better outcomes, better resistance to strategy, etc.), complexity is certainly not a non-issue.
As for chain of custody, again, Australia has solved this problem. Every box is numbered, every seal on the box is numbered. All of them are signed out by electoral officials, witnessed by at least two scrutineers from different candidates. When the boxes are opened this process is repeated.
Do you know how many times ballot boxes have gone missing in an Australian election? Once. In a century. Once. For which the High Court voided the election because there was a minute but non-zero chance that the outcome would've been affected. It was re-run from scratch.
Furthermore, doesn't Australia use it only for House elections where you don't have to compute across a particularly large population?
Your tone is rather dismissive here.
My tone is dismissive because I am tired of being told IRV is "too complex" or "won't work". Because it's simple and it has scaled flawlessly for decades. You just need to believe the experience of other countries is more valid than a theory.
But, I'm with you that IRV is workable (there's ample evidence) and better than the abomination that FPTP is. I guess my point is that AV/SV address these concerns through their simplicity (I find both of them less complex than IRV in both voter UX and result computation) and also that they avoid the unintuitive problems mentioned in the OP ncase link, which can potentially be even worse than the problems of FPTP simply because they are hard for even educated voters to wrap their heads around, whereas the problems of FPTP (vote splitting) are easier to explain (but still bad. this isn't a defense of fptp).