1/3rd octave is nice because it’s easy to implement with a Constant-Q Transform/filterbank. It can also be seen as dividing the spectrum into musical intervals, which can make sense. It’s also easy to alter for different resolutions.
If you’re into more perceptual analysis, you can use the Bark scale which corresponds to the critical bands of your hearing.
Personally I think that using a constant-Q filterbank that has the number of bands as an argument would be the most useful. With the new DSP module it wouldn’t be too difficult to implement.