First, find your own vocal range, if you haven't already. Figure out your limitations on producing powerful vocals without strain. For example, the highest I can go is about F4 before it becomes uncomfortable for me to sing like that.
Next, figure out what scale you want to sing at, and how many octaves you want to cover. Then, figure out how many pitches you actually want to record. The idea of a powerscale is that each pitch will gain in, well, power as it goes higher, to mimic the tendency of songs and singers.
Ritsu Namine: Kire, the most popular example, goes about one and a half octaves [A3 + D4 + G4 + C5], though not on a set scale. Same for Meiji Gahata: Fantasy [A3 + C#4 + E4 + G4 + C5] and Merry Kohaku: Acrobat [A3 + D4 + F4 + G4 + C5], two other notable voicebanks, so if your goal is to imitate them, that's about the range you should look at.
As far as selecting pitches, avoid leaps greater than six halfsteps, as they will not transition well, as well as leaps less than two halfsteps, as they are too similar. Personally, for my powerscales, I use a four pitch single octave with four-halfstep transitions: E3 + G#3 + C4 + E4.
Be sure to alias each sample with the appropriate suffix so that UTAU can read it correctly and that you can use all pitches effectively in covers, ex. "a かE3" or "a か_C4"
Also, it's best to keep your voicebank organized by placing each pitch in a separate folder, appropriately named for what it contains. This will also better organize your oto.ini.