Essentially this HPC idea comes down to a per-program VNC-like system where for example you could run the Blender animation program from a netbook or tablet computer and have that forwarded to a more powerful system, via Wayland. Though with the original proposal, this wouldn't end up being solely the work of Wayland but other components would need to come into play too for all of this to work out. This discussion though has brought up some discussions regarding the serializing of application windows to suspend-and-resume them, etc.
There is a proposal for this year's Google Summer of Code for a student developer to begin work on a remote Wayland protocol, but as of right now, it doesn't appear any student developer has stepped up to the plate.
Another option would be to just run your GTK+ applications from the web-browser beginning with GTK+ 3.2 and using its new Broadway back-end.
Mentioned last night was then work that's already in development by an independent contributor, Trevour Crow. Trevour's work effectively comes down to a "netpipe" driver to Gallium3D that would make it possible to transparently relay drawing commands across a network from one machine to another via Gallium3D. This is being done since the indirect GLX mode with X.Org/Mesa is limited to OpenGL 1.4. Though if fully implemented, this "netpipe" implementation wouldn't just be limited to just graphics but any supported state tracker such as OpenGL ES or even OpenCL.
This Gallium3D netpipe concept sounds faintly similar to VMware's Gallium3D solution for passing Gallium3D commands from a virtual machine to the host machine for native execution on the GPU, but instead the network would sit in between. Efficiently compressing the massive amounts of data though before transfer on the network may be the bigger obstacle.