Well, definitely have a look at the intermediate files, so the problem can be narrowed down to a specific step. I struggle to see how the conversion to .nii.gz could cause issues. Maybe the FSL viewer itself is the problem, but again I’m not sure how that could be explained… So first things first: try and get the .mif output of the (successfully completed) dwi2fod on a computer / system that has mrview, and take a look at the .mif directly in mrview.
Nope, doesn’t help. 
But it’s also unrelated: this is output from the subsequent mtnormalise step. There’s a few explanations for that error, but it’s best to resolve the original problem first. If something is indeed wrong with the .mif output above (but mrview can confirm that or rule it out), then it’s not worth to try and resolve this mtnormalise error at this stage. If the .mif files of all tissue types do look fine however, and you’re still getting this mtnormalise error, we should be looking into that separately.
Already getting ahead of myself anyway though about the separate mtnormalise problem / error: I recall from other posts you’re working on single-shell data (i.e. single non-bzero b-value, and b=0 image(s)), right? Running dwi2response dhollander on that is 100% fine, and in case of single-shell + b=0 data gives you 3 response functions (WM, GM, CSF) for that. Nothing wrong at that stage. But (big but here), running dwi2fod msmt_csd, with 3 tissue types, yet only with single-shell + b=0 data, is not enough. With “not enough”, I mean that you’ll still be able to run it, and get an output, but typically the GM volume will be all zero (apart from small numerical deviations). Putting that near-zero GM volume together with the WM and CSF volumes into mtnormalise though, becomes the problem, and (without going into the technical details why) typically results in that error. Also, even if it wouldn’t result in that error, the output of such an operation is inappropriate. Long story short here: with single-shell + b=0 data, at the moment (without SS3T-CSD being available), you should only perform 2-tissue CSD, for which you can still use e.g. the WM and CSF response functions from dwi2response dhollander. See also the post I linked to in my other reply (that has the long explanations on the response function algo’s and CSD algo’s); and further links in that linked post.
But well, this long paragraph getting ahead of myself is really only relevant at all if the original problem is either solved (or was just a visualisation issue in FSL, or mistake in the conversion to .nii.gz).
I hope that’s all clear. 