Just FYI: my dissertation was largely based on packet-capture results.
The point I started at was Vern Paxson's state-of-the-art in the late
90s. I added additional tuning, and API fixes to libpcap, on top of
that.

I don't know, offhand, of any good papers on this. compared to the
results from the traces themselves, bpf tuning seemed trivial at the
time.  "Increase bpf buffer size" was well-known in the SIGCOMM
bpf-using community at the time.

What timeframe were mjr's results?