Although Andy might have been mistaken in his review that the newer versions are using pure perl, I believe he has good reason. I just benchmarked 1.52 against 2.015, and 1.52 is a little more than twice as fast in both compression and decompression as 2.015.
Compress::Zlib used to be a nice XS wrapper around the zlib library, and it was fast and efficient. Now it has become a front-end to the pure-Perl IO::Compress::Zlib, uses twice the memory, and has almost certainly taken a performance hit. I do not understand the reason for this change, please consider going back to using just zlib.