If you're considering zlib for compression (with existing snappy), you may be better off considering lz4. It's substantially faster at compression and gets very close to snappy in terms of overall ratios. In some basic tests we've done for in-memory compression, lz4 is the only one of the 3 that appears to be capable of feeding data at line speed without recognizable latency.

We get 9:1 compression out of snappy and about 7.5:1 out of lz4. However, our systems compress with lz4 at least twice as fast, in memory.

I'd be happy to run some additional tests... and would love to see lz4 as a compression option everywhere. (lz4 file compression request coming soon)

Relates to: SERVER-3018 and SERVER-25620