I generated a benchmarking script to avoid caching.
For my test I used
50 asyncrounous processes, sending each 30 HAM and 30 SPAM request while
each request was different to avoid cache-effects (DNS and policyd-weigh cache), also the requests where
send over a 100mbit network, DNS requests were send over a 2Mbit uplink:
$ time ./gen-bench-mail.pl
Which means, inclusive network delays (DNS, querying policyd-weight on a remote machine), 3000 Mails were processed in 71
The machine which runs policyd-weight is a Dual P3 1,2 Ghz. On this machine
policyd-weight can handle 2523 Mails/Minute or 3.633.100 Mails/Day.
The test is not an accurate measurement, but it gives you a figure.
Please note, that it is even faster when it can use cached data, i.e. client/domains it has already scored/evaluated.