Replacing the CPU on on one host often puts every system in the rack at risk. Most household systems can stand a loss of a few % of performance with a patched kernel. Server rooms filled with racks and blades, such as a major data center hosts, can mean unscrambling rats' nests of cabling to extract a host, opening it up, edging blocking components out of the way, releasing the heat sinks, replacing the CPU, _replacing the thermal paste_, and re-attaching the heat sink, closing the system up, and testing it. Much like taking your car in for an oil change, this creates a real risk of making mistakes and requiring additional effort or replacement parts. It also creates a risk of failures in the weeks after the CPU replacement, especially if the installer mishandles the thermal paste.
The risk is compounded in environments with poorly configured cabling, such as those shone here: https://www.cepro.com/photo/th...
Completely halting and then cold booting servers is not a zero risk operation. Hosts, or arrays, that have been stable for years will fail to reboot and may even be unrecoverable in an environment where systems are not rebooted regularly and discovered earlier. Mechanical parts, such as fans, and spinning hard drives, are most likely to fail during such a restart. Old clock batteries can expire and fail to set time properly on reboot, old power supplies can fall out of spec and fail to handle start-up voltage requirements, the list of potential problems is extensive.
Replacing CPU's in a production environment can be as great a risk as the security issues of these Intel bugs.