Tensor-valued diffusion encoding enables disentangling isotropic and anisotropic diffusion components. However, its impact on estimating brain microstructural features has only been studied in a handful of parametric models. In this work, we evaluate the Magic DIAMOND model, that allows characterization of crossing fascicles and assessment of diffusivities for each, using combinations of linear, planar and spherical encodings in vivo. Building statistics through stratified bootstrap, we show that spherical encoding substantially increases the variance in estimated parameters and should be avoided. Planar encoding, on the other hand, did not offer clear improvement or worsening within our current acquisition scheme and setup.