While this theory holds for a single workstation, it doesn't work so much when you throw VMs into the mix. I have physical servers hosting dozens of virtual machines. This means I can overcommit RAM for two reasons:
1. Some machines are used infrequently, and are better off sitting in swap, leaving RAM space available for more frequently used machines.
2. Several of them quite often run identical operating systems, and memory pages can be shared between them with an appropriately configured hypervisor.
Whereas before, it was inactive applications that got swapped out to disk, I'm finding more and more these days that it's inactive servers that get swapped out. Storing the swap on SSD's has significantly reduced the time to live on these machines. And while a few gigs of RAM is cheap for a workstation, 64 Gb of fault-tolerant server-class RAM... not so much.
Using SSD's to house swap memory, I've found I can maintain the same level of performance while requiring less RAM.