Your suggestion about doing temp calibration with just Lum and then filter offsets later is a viable option.
If your temps drop a lot overnight (as mine do), I think it's best to record three numbers for each successful focus: filter, position, and temperature.
Take a lot of data points (dozens), particularly if you have a lot of scatter for the NB filters. For me this takes at least all night, sometimes several nights.
Then after the fact create a spreadsheet with the three columns (filter, position, temp) and plot focus position against temperature for each filter.
You should get fairly linear behavior between focus position and temp, with the slope of each filter line being very similar to each other (parallel lines). The lines for each filter should be offset from one another.
Run a linear regression for each filter, to get a slope and intercept.
The average of the slopes is your temperature coefficient.
Subtract the y-intercept of each filter's line from the y-intercept of the Lum filter. These are your focuser offsets.
Here's a program in R that does all this, and an example of the output for my system.