What is the minimum number of peaks (or smallest spectral width in your case) to get at similar result with reasonable confidence?
I think David Boyd has also used this technique.
Yes is it is a standard technique used to measure radial velocities. I first came across it in spectra from the CfA Digital Speedometer eg
and was impressed by how noisy spectra could be and still give good RV measurements
Buil used it for exoplanet measurements where he estimates a precision of 20 m/s on 100 SNR echelle spectra, cross correlating 255 lines.
I think it is quite difficult to estimate the uncertainty in general though as it depends for example how narrow the lines are as the digital Speedometer paper shows
In this case I suspect the accuracy is probably limited by systematic effects rather than the precision of the cross correlation. One obvious problem with slit spectrographs is the precise position of the target on the slit. If it is slightly offset this can skew all the line profiles slightly which can have a big effect when trying to measure to small fractions of the resolution. (1 km/s is ~1/20 of the spectrograph resolution). Here this could affect the asteroid spectrum but not the sky spectrum as that is an extended source . The guiding is quite poor with my setup though which probably helps average out this effect. Fibre fed spectrographs like the one Buil used don't suffer from this as the fibre scrambles the profile from the slit. (Buil also used a mechanical scrambler (oscillating fans!) in the fibre feed t oget down to his very high precision.
I've not seen cross correlation used to compare line widths. I guess using more lines would increase the discrimination but I am not sure how you would go about measuring it. Similarly there must be techniques around to disentangle composite spectra from binaries, based on template fitting but it is not an area I am familiar with.