Provable privacy based on 'anonymized' published data sets that resists third-party information and linkage is really hard -- arguably impossible in general. When people claim 'anonymized' data, I tend to not be fully convinced.
I completely agree with you on the difficulty of truly anonymizing data.
I do wonder if there any holes in this idea though: What if they released the data the number of cars per hour per segment of road (a segment meaning a continuous piece of road between two intersections) rounded to the nearest multiple of 10?