I have no direct experience with this exact situation, so take what I say with the appropriate measure of salt.
Somewhat surprisingly, most of the heating done by these systems is by radiation, not by free convection. Infrared light is just as good at shining downward as it is at shining upward, so the "heat rises" comment, while true, is also a bit misleading.
Insulating per se is not required, unlike in a situation where you have tubing mounted UNDER a wooden floor and hence you need very high temperatures for the floors to work at all. Insulation is less necessary with a really excellent thermal coupling between the tubing and the floor surface you want to heat (i.e. what you'll get when you embed the tubes in a floor topping such as lightweight concrete or gyp-crete). The slabs are hollow so the top-to-bottom thermal coupling isn't going to be perfect even without insuation. Insulation becomes more necessary any time the client wants to top your gyp-crete covering with something of poor thermal conductivity, i.e. hardwood or cork etc.
Insulating the underside of the top floor will definitely make the system easier to control to the client's satisfaction on a floor by floor basis- the top floor frequently having the MOST heating demand when you actually run the calcs- but so would the inclusion of some radiant walls or a couple flat panel rads in the design.
The big challenge with these high mass radiative heating systems is temperature overshoot during spring and fall. These systems are not good at tracking setpoint changes. A few faster-responding elements such as rads in the design, plus a good outdoor reset temeprature controller, help a great deal with making these systems work satisfactorally.