Sounds like you are thinking about things right so far. In our experience, ACBC works reasonably well when each level appears a minimum of 2x per respondent. You could obtain more precision by having each level appear 3x per person. But, the extra precision at the individual level could be made up by reducing the individual respondent burden (as you've done) but having a reasonably large sample size.
You've said nothing about sample size. For studies about populations of 5000 potential customers or more, you generally want to have 300 to 500 respondents, I would think. Another rule of thumb is to run the Test Design report in ACBC with the sample size you are expecting to achieve...then checking to see if the standard errors for each of the part-worth utilities is under 0.05.
