Hi, Robin-

The blurring "default" is 4mm for no deep reason-- that is, it is not necessarily a "recommended" value, it is just a place holder. Note that for typical EPI voxel sizes with edges of ~3-4mm per side, this would result in very minimal smoothing. If you want noticeable smoothing, maybe having a value 2x your voxel edge length would be more noticeable.

As to the pros/cons of smoothing in terms of strengthening some signals and potentially washing out some small regions... yep, life is full of tradeoffs! But note that Bob developed the ETAC methodology *specifically* to address+reduce the arbitrariness of smoothing values on final cluster results; see this paper:
[www.biorxiv.org]
[www.liebertpub.com]
This methodology would involve *not* smoothing your initial data, and then using different blur sizes later on, and combining all results "equitably" (while also combining results across different semi-arbitrary voxelwise threshold values, too!).

--pt