I went to the library and took a look in Johnson and Jasik's "Antenna
Engineering Handbook," where a good explanation of the shielded loop antenna
is provided. In brief, what happens is that the incident magnetic field
induces a current around the _outside_ of the shield and -- due to the slit
in the shield -- the current is then forced to flow "around the edge" and
onto the _inside_ of the shield through whatever impedance it sees at the
slit. If you're using a piece of coax where the center conductor has been
shorted to the shield at the location of the slit, the impedance looking
into the slit is the load impedance 'transformed' back the length of the
transmission line (i.e., assuming a matched load, it would be the same as
the transmission line impedance). If you're using a coil of insulated wire
inside, e.g., a copper pipe, the slit impedance is pretty much a short
circuit (well, a small inductance, actually -- the self inductance of the
inside loop) so all the current induced on the outside is 'mirrored' into
the inside. This then couples into the coil of insulated wire just as if
the shield weren't there. Some small loss occurs due to the loop inductance
and the finite conductivity of the shield. (I've found web sites with
measured results showing that shielded loops do have readily measurable loss
over unshielded loops.)

An interesting result of the above is that the polarity of the voltage
coming out of a shielded loop should be opposite that of what's coming out
of an unshielded loop. (Because in the case of the shielded loop, as the
current flows over the edge of the slit it's reversing direction.) It would
be fun to try this experimentally... (I'm thinking something like building
another coil and discharging a capacitor through it to induce spikes in the
antenna.)

The explanation on web sites about the shield acting something like an
electric dipole doesn't really fly, IMO.

---Joel