This came up in another thread recently:
We could potentially get a big performance boost at minimal file size cost by changing the default gz compression level to 1. In some small-scale tests I did a few months ago, there was a major time cost between high and low settings, for only a tiny compression improvement (e.g. 90-> 26 MB at level 9 in 34s, vs 90->27 MB at level 1 in <1s). This was only a small test on a few brain MR datasets, but it seems reasonable.