Sockets are also a problem some times ... the first product I ever worked on used 68000 CPUs in zero insertion force sockets. (Yes, this was more than 25 years ago.) The product was a data network switch and was the size of a large refrigerator. The first cabinet we shipped arrived at the customer site and the technician could not get it to start up. Then he called back and asked why there were all of these long chips with gold pins in the bottom of the cabinet. Almost all of the CPUs vibrated out of their sockets during the ride on the shipping truck. After that, we started to solder them in. The original reason for the sockets was to make repair easier, but they caused way more problems than they solved and the CPUs were very reliable.
So, it is easily possible that Apple was having more DOA minis caused by issues with the sockets then it had with truly dead CPU chips. At some point, there is a break even where the number of CPU repairs on the socketed CPUs will exceed the number of main boards you would have had to pitch if they were soldered in. (And add the savings of not paying for the socket.) I sincerely doubt we will ever know the real reason for the change. In general though, true CPU chip failures are rare ... the only real gotchas are when Intel has a brain fart and ships CPUs with bad microcode or other defects.