I appreciate everyone's attention to this. It turns out the culprit was rapportd trying to talk to an ipad. Turning off the ipad (and letting rapportd settle down, I guess it persists trying to call out for a while if the network is down?) for several minutes made the problem go away.
In case anyone runs across this, I was able to correlate the spikes by looking at `iftop` output (which tracked to mdns.mcast.net going to the ipad IP), `netstat` to see what was trying to get to the IP, and `sudo fs_usage -w -f network rapportd` to confirm the bursts of activity
There's clearly something wrong with the packet accounting somewhere between rapportd and Activity Monitor for some reason, but local network traffic was the cause. Given the large numbers, I wouldn't be surprised if it was related to some bitwise operation error (max 64bit unsigned is 18e18 and it was showing 17e18, not close enough for me to believe it was an underflow).