This was one of the same concerns I had at the beginning of our research into Summed Pricing and ACBC. About six years ago we conducted an ACBC experiment in which half of the respondents saw level prices for a few of our attributes that were very different from the other half. We wanted to know if setting the prices different for level prices in Summed Pricing would have a large impact upon the final estimated utilities.
For that experiment, we tried to be quite dramatic in our manipulation of prices between respondent groups. One group of respondents (n~200) saw price levels for a brand that were in complete contrast to the other respondents (n~200). To one group of respondents, we attached the highest relative incremental price to a brand and to the other group of respondents we attached the second-to-lowest incremental price to that same brand. We did other sorts of dramatic manipulations between respondents on two other attributes.
Then, we fielded the studies and estimated the utilities to compare the results between the groups. Although we found a few statistically significant differences, the differences were not especially large...and not exactly predictable based on the manipulations we did. We thought that putting higher prices on certain levels could potentially signal to respondents that such levels should be worth more--leading to higher part-worth utilities. But, the effects were inconsistent across our manipulations.
In the end, we came away from that experiment with the impression that the final utilities were quite robust to even quite perverse choices of prices to assign to levels within the Summed Pricing capabilities of ACBC. It seems that if the researcher and client can choose loosely reasonable price increments for use in ACBC Summed Pricing, it should convey benefits to the rest of the experiment in terms of tending to avoid product combinations with the best features at the lowest prices and the worst features at the best prices. It also makes sure that the BYO responses involve tradeoffs between enhanced attribute levels and prices. This should make the interview more realistic and meaningful to respondents.