Alternatively, a brief and by no means comprehensive list of things I've found to be better in Renderman versus Cycles:
* Bidirectional pathtracing and VCM for fast, accurate caustics and SDS (specular-diffuse-specular) illumination without fireflies. [0][1]
* Extremely efficient subdivision surfaces and displacement. In Cycles, there is a performance penalty for subdivs and displacement. In Renderman, there is almost zero overhead. [2][3]
* Significantly faster subsurface scattering [4][5]
* Support for OpenVDB volumes, and much faster volume rendering/scattering/etc. [6][7]
* Importance sampling for emissive effects, such as explosions and flames. Think candles with actual fire that don't firefly! [8]
* Importance sampling of HDR maps (basically, prevents fireflies when using HDR maps) [9]
* Much better memory efficiency. Renderman is designed to handle literally hundreds of gigs of stuff in memory going in and out of core all the time. Don't have a link for this one, since this comes mostly from experience with both renderers and having used Renderman in a large production studio.
* Disney "principled" BSDF [10]
* Arbitrary geometry lights, with importance sampling [11]
* Better/faster hair rendering and hair shaders [12][13][14]
* The Denoiser is PRMan 20 is legitimately magic. [15][16]
Generally, there isn't necessarily any particular feature that Renderman has that Cycles is outright missing, but almost everything in Renderman is a lot faster and more optimized (as seen in the link list below, a lot of the research on making stuff like subsurface scattering and hair and whatnot extremely fast comes from Pixar and the Renderman devs/researchers in the first place). On the flip side, Cycles has a usable GPU mode, whereas I doubt Renderman will ever get a GPU mode anytime soon. Cycles is really pretty great, but having a volunteer team versus a dedicated, paid team that has to support not just Pixar's production studio but also a ton of large VFX houses produces a much faster development pace and a much higher incentive to optimize the crap out of every inch of Renderman.
Disclaimer: I now work for a different production renderer team at a different studio (some consider us to be a rival to Renderman, we don't see it quite that way), but I've worked at Pixar before.
[0] http://renderman.pixar.com/resources/current/RenderMan/PxrVC...
[1] http://cgg.mff.cuni.cz/~jaroslav/papers/2012-vcm/2012-vcm-pa...
[2] http://renderman.pixar.com/view/displacements
[3] http://graphics.pixar.com/opensubdiv/docs/intro.html
[4] http://graphics.pixar.com/library/ApproxBSSRDF/paper.pdf
[5] http://graphics.pixar.com/library/PhotonBeamDiffusion/paper....
[7] http://renderman.pixar.com/resources/current/RenderMan/rfmOp...
[8] http://graphics.pixar.com/library/MISEmissive/paper.pdf
[9] http://renderman.pixar.com/resources/current/RenderMan/PxrSt...
[10] http://renderman.pixar.com/resources/current/RenderMan/PxrDi...
[11] http://renderman.pixar.com/resources/current/RenderMan/risLi...
[12] http://renderman.pixar.com/resources/current/RenderMan/PxrMa...
[13] http://graphics.pixar.com/library/DataDrivenHairScattering/p...
[14] http://graphics.pixar.com/library/ImportanceSamplingHair/pap...
[15] http://renderman.pixar.com/resources/current/RenderMan/risDe...