The previous compression algorithm does not compress everything it can
in cases where there are domain names in the data section of a RR, e.g.
when the answer contains
bar.example.com CNAME foo.example.com
foo.example.com A 192.0.0.2
and foo.example.com was not encountered previously, then the second RR
leaves the 'foo' part uncompressed.
The improvement resolves some of these cases by comparing the owner of a new RRset to be written
with the last written dname instead of QNAME. Also, the effectivity was slightly increased
according to benchmarks.