If there was close contact, if the 90% rate is accurate, and if occurrence is independent in related individuals, then you would expect 0.10 * 0.10 = 1% of contacts with 2 potentially vulnerable people to result in neither person infected.
1% sounds rare, but rare events happen all the time, and 1% isn't even particularly rare. If you know 100 families, you'd expect this outcome to happen on average in 1 of them.
That's not very unusual and is clearly plausible just from the information you have at hand. As @DeNovo mentioned in a comment, it is also likely that the spread is not independent, because the children share several characteristics: they are related, so they share:
- any genetic component to vulnerability
- any characteristics of the father's illness such as the level of virus replicating in the father's lungs
- perhaps the level of actual contact with the father and how well he may be effectively quarantined from the others
Those factors could make the joint probability across the two children closer towards the 10% rate for a single individual: once you know if the first child is infected or not infected, you can make a better guess about the second child based on all the possible shared characteristics of the children or the infected person.