What is the smallest cluster size where using "Percentage of resources reserved as failver capacity" makes sense? If I have 3 hosts in a cluster and I need N+1 redundancy using this policy, should I start with 34% reserved?
Depends on your needs actually, as Percentage of resources reserved is the recommendation you would take 100% and divide it by your number of hosts to get n+1.
Keep in mind that you might still get issues that way though:
3 ESXi, 4GB RAM each
1 VM 2GB reservation running on ESXi-A
1 VM 2GB reservation running on ESXi-B
1 VM 3GB reservation running on ESXi-C
Cluster has a total of 12GB RAM with 7GB used, n+1 would be 66% of 12GB RAM which is 8GB, so you would not violate admission control. But what happens when ESXi-C fails? Due the large reservation you will not be able to power on the VM on either of the failover hosts.