This drove me up the wall trying to figure it out, but loading jpg's or tiff's into a Cairo surface with Python really isn't all that hard.

The trick is to load it into a gtk.gdk.pixbuf first. here's an example:

pixbuf = gtk.gdk.pixbuf_new_from_file(filename)
x = pixbuf.get_width()
y = pixbuf.get_height()
''' create a new cairo surface to place the image on '''
surface = cairo.ImageSurface(0,x,y)
''' create a context to the new surface '''
ct = cairo.Context(surface)
''' create a GDK formatted Cairo context to the new Cairo native context '''
ct2 = gtk.gdk.CairoContext(ct)
''' draw from the pixbuf to the new surface '''
ct2.set_source_pixbuf(pixbuf,0,0)
ct2.paint()
''' surface now contains the image in a Cairo surface '''

Of course you could bypass stamping it onto a Cairo surface entirely and just use the original pixmap for most things, but for doing affine translations I needed an actual surface and not a context to one. Otherwise you have to remember what is accessible through gdk and what is a native Cairo surface.

With a little bit of pipe magic, you can even do image manipulation externally before loading it, and never have to create a temp file:

import subprocess
''' Load the pixbuf from the embedded thumbnail in a RAW file using dcraw '''
p1 = subprocess.Popen(["dcraw","-e","-c",filename],stdout=subprocess.PIPE)
myloader = gtk.gdk.PixbufLoader()
myloader.write(p1.stdout.read())
myloader.close()
pixbuf = myloader.get_pixbuf()

Then pass the newly created pixbuf (actually a JPG created from dcraw) into the first swatch of code, and you've gone from the embedded thumbnail inside a RAW camera image into a Cairo surface.