It’s one of those “can’t hurt, might help” kinds of suggestions.
While zigbee mesh is self healing, it doesn’t mean quite the same thing as the full zigbee heal, Which is what happens when the coordinator (hub) is powered down.
“Self healing” in this context simply Means that if a preferred message route is unavailable, the sending node will try an alternate route. There are multiple steps involved in all of that, but that’s really all it means. No human intervention is required in order to get a message to take a different path. And this process, as you guessed, is very quick, usually less than a minute.
The problem comes in educating the nodes as to what potential paths are available, and the depth of the network at each repeater (Different model devices support different numbers of children, so if a parent that could support five children goes bad and the only alternative available only supports three children, you can be left with “orphans“. )
If you rely solely on zigbee self healing, It’s quite possible to end up with “orphan nodes” which know they belong to this network, but which have no parent to repeat messages for them, even though there are parents with available child slots. 
Also, in some cases when potential parents were added to the network after the children already existed, those parents end up never being considered as possible substitutions. This just gets complicated.
When you temporarily remove the hub (the coordinator) from a zigbee network, The other devices assume that the network is being physically moved and that none of the previous paths can be relied upon. Consequently, when the coordinator comes back, everybody starts over in building their preferred paths. This will generally pick up any orphans and also improve efficiency if new parents are available who would be a better first choice than the old ones.
So in this sense “healing“ (sometimes called “forced healing”) Is a much more thorough process and allows every end device to consider every possible repeater as a parent candidate. This process can take much longer than a self heal, Sometimes hours, to complete, so field techs usually don’t check for the full results until the next day.
Self-healing just occurs when a sender tries to get a message through, the message doesn’t get through, and the sender asks itself “hmmm, do I know of any other paths to that destination?“ self-healing does not normally assign new parents to orphan devices.
So self-healing is a good thing, and helps keep the network operating When there are temporary problems along an individual message route such as local interference. But a full heal Essentially rebuilds the whole network and causes more efficient parent/child assignment, as well as picking up any orphans.
I’m sorry, I don’t know of any really simple explanation for this, but here’s A good technical paper which describes the difference between what happens when the coordinator (hub) goes off-line versus what happens when an individual repeater like a light switch goes off-line.
The one thing to note is that in their example it’s a sensor net which is choosing a new coordinator. In the zigbee home automation profile, which is what smartthings uses, there’s only one potential coordinator, a hub, and it’s either there or it isn’t. But the distinctions between the types of failures are similar, particularly with regard to orphan nodes.
The following paper was published in a peer reviewed journal, the International Journal of Computer Science Engineering and Technology (IJCSET)
http://www.ijcset.net/docs/Volumes/volume3issue3/ijcset2013030309.pdf
This is the key distinction
When a device is found orphaned {during a forced heal}, a realignment or a channel re-scan process will be invoked.
That will happen only when the coordinator (hub) is powered off for at least 15 minutes and then on again.It does not happen in self healing.
