This is most likely related to rounding errors causing different groupings of shells (see here). The current dev branch has a reduced threshold of 80 and you should not see this issue with your data.
Alternatively, I think you could manually increase the b-values corresponding to the b-100 shell (or the vector norm, using our b-value scaling) by a small amount to counteract rounding issues. However, the cleaner solution is to use the development branch (git checkout dev; git pull; ./configure && ./build).