Although I have not used this software in quite some time, you can download an application that does the power calculation at the following website:
As a general rule of thumb, minimal number of participants for an HLM design is 25 groups with 25 participants per group. A third level would obviously require a similar number of dimensions, meaning that your sample requirement will be very large.
Here are a few references to get you going:
Paterson, L, & Goldstein, H (1991). New statistical methods for analysing social structures: an introduction to multilevel models. British Educational Research Journal, 17, 387-393.
Snijders, T., & Bosker, R. (2000). Multilevel analysis. Thousand Oaks, CA: Sage Publications.