Cheap, no: $5.39 in single quantities.A cheap, simple and accurate way is also the Dallas DS18B20 (10 bit capability), that I already mentioned.
Simple, maybe: If code libraries are available for the one-wire interface, then yes. If not, then no.
Accurate: 0.5C (-10 to 85). The thermistor is still better.
One thing that bugs me about the part is the 1.5mA active current. During the active times, that current will cause the part to dissipate 7.5mW (at 5V). This will cause an self-induced temperature error dependent upon the thermal characteristics of the package. There is no data for this on the datasheet, but given a typical TO-92 thermal resistance of 80 to 310 C/W, an increase of 0.6 to 2.3C can be expected during those times.
They also seem to give no thermal time-constant data (I only quickly perused the datasheet!). This is oftentimes important, especially if you are trying to design an active controller.
Final objections: 3 wires vs. 2. Must observe polarity/pinout. Not easy making a temp probe from a TO-92 package (intended for use as PCB mount). Lead length limited by digital interface. Resolution is limited to manufactures design. I've run the thermistors to 0.01C resolution (and better!).