If you had said, "I don't think making sure that things don't happen instantly is explicitly part of their grand plan" then I would dispute that statement. (Yes, making sure things don't happen instantly is explicitly/implicitly part of their grand plan.)
They want users to have an "experience" not just click a button or press a key and instantly see the result. It is very explicitly part of their plan, and if the algorithms don't slow things down enough, they actually request animations or loading screens that do so.
Some of it may be implicit, rather than explicit. For example, a front-end javascript framework might actually produce pixel for pixel the exact same result as a simple HTML file generated more quickly on the back-end might do, but could still be a preferable "experience".
All this could be tested explicitly. For example two groups of PM's could evaluate web sites, where one included Working... screens where a javascript framework slowly renders something client-side, and the other could have pixel-for-pixel the same thing served from the server, using less bandwidth than it took to even transfer the javascript files over, so that the result is served much, much faster.
I hereby formally predict that some PM's will prefer the slower version for the exact and only reason that it is a slower user experience, and that this could be verified with a double-blind and carefully controlled experiment. I've put my claim out there, and it is easily falsifiable as well as has explanatory power in terms of the software we actually see.
Very simply: some PM's prefer a slower user experience. (I claim.)
I highly doubt the ATM maker put in sleep calls to intentionally slow it down -- no one is going to be surprised if it takes 1 second instead of 10 seconds to scan a check.
The bank could optimize for time, but it's not their time, it's their customer's time and they are already saving so much money over a teller transaction, they likely don't care about saving customers a few seconds.
When you do get to use it, using performant software is a really great experience, so how to we train users to appreciate this?
Ditto BitTorrent clients like uTorrent and Transmission; the first generation of BitTorrent clients were written in Python or Java, and felt like it. When users started complaining about how they would lock up your machine, somebody actually bothered to write one in highly-optimized native Win32 C, with no frameworks.
You have to understand what the customer is optimizing for though. Someone standing at an ATM is optimizing for confidence, not speed; they already took the time to drive to the ATM, so depositing their check in 1 second vs. 10 seconds doesn't matter when the whole trip takes them 10 minutes. Someone who's responsible for depositing 1000s of checks for a major business, using an office scanner, cares a lot. But I really doubt that industrial-strength check cashing apps put a sleep() statement when scanning the checks.