Digit Span Tests
6 is the average
The adult average for the famous (auditory) digit span test is just above 6. The average for visual digit span test is roughly the same.
Around 80% of adult population will score between 5-8 in such tests, and people scoring 4 or below will be suspected of some cognitive impairment.
4 is the boundary
This means that most adults should be able to recall a chunk of 4 digits (which one party will have to do over the phone).
But do bear in mind that both children and the elderly score lower in these tests. So do people with cognitive impairments, like dyslexia.
Just because you can doesn't mean it's easy
Most of us can calculate 7*9/2, but would find it easier to calculate 14/2.
Similarly, it is easier to recall 3 digits over 4, as both these graphs demonstrate (source):
Group count/items tradeoff
So with 4 being safe and 3 being easier, it's really a question of striking a balance between the amount of digits per cluster and the amount of clusters. It is worth remembering that as people read these numbers out loud, they have to keep track of which group is being read - a seemingly simple task, but even simple tasks can be challenging given a particular context.
3 per cluster
For 14 digits, you'll end up with:
XXX XXX XXX XXX XX
4 per cluster
Will give you:
XXXX XXXX XXXX XX
5-3 vs 4-4
Although you can't regard in the same way the amount of digits to the amount of clusters, a 4-4 system seems more balanced than a 5-3 one.
5 chunks would be perceived as more effortful compared to 4, particularly as most people will have no problem with a group made of 4 digits.
What's more, when one person says it and another has to write it, the back and forth toggling which happens between each group takes time. By way of analogy, if you have to move 12 chairs between two rooms, would you prefer to move 3 at a time (4 rounds), or 4 at a time (3 rounds)?
The only reason to prefer a 5-3 system is accessibility (which would include children and the elderly). But in this specific case such claim is based on reasoning - I wouldn't pick the 5-3 system without further research into this specific issue with the cognitively impaired.