DNeg are still using Mantra (and Clarisse) for a fair amount of stuff (Ant-man is their first production using just RIS in PRMan 19/20), Framestore are solidly Arnold now, and Weta are using their own PRMan clone (but a path-tracer) called Manuka.
PRMan's volume support in RIS is still pretty poor - even in 20 - they're still using generic Woodcock tracking which just doesn't work well, as the max extinction coefficient is global to the whole volume, so is very inefficient as you can't localise the step distance efficiently, or importance sample the density integration. Similarly, in RIS, the importance sampling for emissive volumes is non-existant (unless you write your own interior integrator).
Hair rendering in RIS is only fast-ish because they tesselate to triangles (based on the shading rate), which means stupid memory usage.
Pixar's (Disney's) denoiser doesn't really work with hair.
PRMan's volume support in RIS is still pretty poor - even in 20 - they're still using generic Woodcock tracking which just doesn't work well, as the max extinction coefficient is global to the whole volume, so is very inefficient as you can't localise the step distance efficiently, or importance sample the density integration. Similarly, in RIS, the importance sampling for emissive volumes is non-existant (unless you write your own interior integrator).
Hair rendering in RIS is only fast-ish because they tesselate to triangles (based on the shading rate), which means stupid memory usage.
Pixar's (Disney's) denoiser doesn't really work with hair.