Yes, thanks for your reports. We've seen this and are currently checking where it comes from. The upcoming version embedded
dnsmasq has still a few bugs as well and Pi-hole users already helped us identifying a quite severe bug which we reported upstream already. It is a subtle bug that manifests in the DNS server believing the upstream DNS server munged a domain name. As this also prevents domains from entering the cache at all (without logging that this happens), it may be that the issue is more widespread and more domains are just never really added to the cache, resulting in queries being incorrectly reported as cached whereas we had to ask upstream. This seems to be the best explanation for the observed behavior.
Could you provide a few examples of such domains from your
/var/log/pihole.log file? Even when the timestamps are not millisecond accurate, it should show us what happened to the queries.