There must be a typo in the colour components values, because this corresponds to a pink, not to a dark grey...
I cannot reproduce this issue, here, and still see the sphere, whether in mouse look or not... Albeit not always with the same color/transparency (likely due to some occlusion culling issue).
This could be due to the very very close "camdrawalphamin:0.6" setting (60cm from your avatar eyes), which might be "too close" depending on what head attachment you are wearing (bounding box too large ?), allowing to "cheat" through...
This is the expected behaviour, to plug some cheating cases related to face selection highlights that would otherwise allow to see through... See the corresponding code, lines 7632 to 7640 of linden/indra/newview/mkrlinterface.cpp.
Likely in the Windlight renderer days... That code was designed for the WL renderer, got temporarily broken with the EE renderer, then "repaired".
Not sure if it was broken with EE, but you will notice that water surfaces also get through the spheres, unoccluded...
The problem is that the restriction spheres are dealt with in three different places (*), so to deal with the various rendering passes corresponding to renderer settings (mainly ALM/forward). The good news is that the PBR renderer seems to like this RLV code better (no varying sphere darkness issue).
I will have a look (especially for the occlusion issue), but cannot guarantee I will find a solution (this code is fragile and hacky).
(*) lldrawpoolavatar.cpp for ALM, llvoavatar.cpp for forward rendering, and llviewerdisplay.cpp, likely for selection cases where the former two places are not "hit", and where the sphere is rendered opaque.