I added that word because at least with non-bespoke backups, you have probably thousands of users each day testing restoration out of necessity, and word might get around if the method failed to restore. But nevertheless, one should still test even then.
I always see this advice but...how do you even do that without having an entire additional set of disks to restore too?
You can't restore to production obviously, as aside from the downtime if the test fails you've just destroyed your good copy and proved the other copy is also bad.
About the best I can think of is restoring a small part of the set as a sample, which isn't really testing the whole thing.