The beauty of the distributed nature of the ‘chunks’ and hence the fast rebuild times (15mins) mean that as soon as one mirrored copy has been lost (single drive failure) a new second copy is made very soon afterwards. Most dual drive failures occur because of extra stress put on them due to the rebuild operation being performed. When this is targeted at the same disks that are already in a degraded state, recipe for disaster. The benefits are two fold due to the distributed nature.

a) you don’t put a much stress on any one set of drives (so are less likely to cause the second failure)

b) as you suggested with a large enough distribution of drives the changes of a second drive failure causing both mirrored copies to be lost are less.

I’m looking at the lower level architecture and still trying to get my head round some of it too!