Genshi Compiler allows for rendering your Genshi template to Python source code. You can save the code as a Python module or compile it into a directly usable module object in memory. Just call the render function on the module with your template parameters to render the whole template or any of your template functions to render those fragments separately.
According to my initial benchmarks the rendering speed is typically ~40x faster than doing the same using Genshi. There is a cost of this speedup, certainly. Some of Genshi's dynamic features are not available, most notably anything that depends on a template loader (xi:include), the XML element tree representation (py:match) or the token stream (filters).