[kamocat]

This one takes more testing than a premade sensor, but you could use a small buzzer and look at the acoustic response of the mold, similar to how SONAR works. However, instead of looking for a peak in the response, you'd be looking at the waveform amplitude over time and comparing it to the empty mold. (Make sure to cut off the reading fairly quickly so you don't record data from objects outside the mold).
This should be an inexpensive option, but it is only appropriate if there will not be changes to the mold.

Alternately, you could look at the photo-reflectivity of the piece in the IR or ultraviolet range, and detect its presence with a change in that value.

Have you considered a probe with a microswitch?