At Google my colleague developed a simple test that made vibration effects very obvious (this was in 2008). In a server with a bunch of disks, do random reads or writes on all but one disk. On the final disk, do sequential reads. Back then you could get on the order of 100MB/s sequential read speed on the outer edge of a hard drive in ideal conditions. Under this test performance would dive to something like 10MB/s or worse. All of the seeking would cause vibration, which caused misreads on the disk under test. When a bad read happens a hard drive has to wait until the next rotation to try again. We eventually fixed this with better damping.