Gallium3D's LLVMpipe driver allows for systems without a proper graphics processor or hardware driver to run OpenGL over the CPU as a software rasterizer. LLVMpipe takes advantage of LLVM to speed up the experience compared to Mesa's ancient (and useless for end-users) "swrast" driver or Gallium3D's Softpipe by leveraging multiple CPU cores and modern instruction set extensions like SSE3/SSE4 and AVX. LLVMpipe is still too slow for proper OpenGL gaming even with a modern high-end CPU, but it does a decent job for handling a composited desktop if you have a beefy systems.
Ubuntu and other desktop/distribution vendors are fond of the approach of using LLVMpipe for software acceleration since they don't need to maintain multiple versions of their desktop (i.e. the GNOME fall-back mode or Unity 2D) by just forcing users to run LLVMpipe on the CPU. This is mostly fine in the x86 space if your PC is within a few years old, but Unity-over-LLVMpipe on ARM is a complete mess and we are just days away from the Ubuntu 12.10 release. LLVMpipe isn't optimized for ARM (no NEON generation within LLVMpipe or other optimizations) and the current-generation SoCs just don't pack too much of a performance punch.
If you are running on vintage x86 hardware, using LLVMpipe for the desktop could also be troubling or if your Linux desktop is contained within a (non-VMware non-VirtualBox) virtual machine where there isn't 3D hardware acceleration and limited resources. LLVMpipe also does well for x86_64 Linux while it's much slower in the 32-bit space.
So while I have been testing out the Gallium3D LLVMpipe driver for years since its infancy when it was still a very young driver and not widely used or mentioned in any mainstream context, I'm not even entirely fond of the default Linux fall-back for desktops to be using LLVMpipe instead of a non-composited desktop that doesn't require OpenGL. In terms of using LLVMpipe rather than the default software rasterizer as was the case for most Linux distributions until recently, that at least is great since LLVMpipe is half-way useful compared to Softpipe/swrast, even though its performance isn't still up to scratch (it falls short of some Windows-based software rasterizers).
Anyhow, the good news today about LLVMpipe on the desktop is that Martin Gräßlin, the well known KWin window manager maintainer for KDE, has written a blog post entitled To llvmpipe or Not? Simply put, he doesn't have any plans to use LLVMpipe as a default software fall-back when KWin detects that no GPU/driver is supported. Instead, users can fall-back to using KWin without compositing or to use the X Render acceleration method assuming you have an X.Org driver that can handle the X Render extension.
Martin wrote, "I do not consider [LLVMpipe] as a solution for the KDE Plasma Workspaces. I think we have better solutions for fallback (XRender or no compositing)...If your hardware is too old to provide the requirements for OpenGL based compositing, it is quite likely that your CPU is also rather old. We can assume a single-core CPU of either the Pentium era or an early Atom. You don’t want your CPU which is not suited for this task to spend all the time rendering the desktop. If I enable llvmpipe for KWin on my system (quad core) it puts about one core under full steam."
Martin additionally writes, "Overall we see that llvmpipe as a backend for OpenGL based compositing is actually not needed in the case of the KDE Plasma Workspaces. We have working fallbacks like XRender or no compositing at all and those do not put the system under full-steam. Given that I am not considering to turn llvmpipe into an option for KWin. I think that most users who would need llvmpipe are better suited with XRender based compositing even if a few effects are missing then. For me as a developer on the other hand llvmpipe is a very interesting target as it can help to optimize our rendering stack."