First of all, the old adage that "bencharks are meaningless numbers" has a lot of truth to it. For most, if not all, of the real-world situtations out there, BIND, DjbDNS, and MaraDNS provide more than enough speed.
Men, being men, though, like to use benchmarks to time meaningless DNS queries, giving meaningless information, which then they write wordy documents about, so that they can pretend they have useful information, when, in truth, they have learned nothing.
The benchmark performed here was done with the "benchmark" tool supplied with MaraDNS. I sent a series of 50,000 queries over the loopback interface to DjbDNS 1.04, BIND 4.9.7, BIND 8.3.2, BIND 9.1.1rc6, and MaraDNS 0.5.09. Here is how long the queries took, in seconds, to process on a Dell poweredge computer with a single 800MHZ Pentium III CPU, running Linux 2.2.19:
maradns-0.5.09: 3.46 bind-8.2.3: 6.34 bind-9.1.1rc6: 6.40 bind-4.9.7: 6.48 djbdns-1.04: 9.95
<Thumping chest in pride>As you can see, MaraDNS is nearly three times as fast as DjbDNS, and a hair under two times as fast as BIND.</thumping chest in pride>
These numbers, of course, do not tell the entire story. TinyDNS uses a disk-based database, which allows it to do some things MaraDNS can not do, such as:
BIND has many capabilities that MaraDNS does not have, such as dynamic DNS and signed DNS records.