Larger sample sizes increase the possibility of Type II errors. Too small versus too large is a common line researchers walk. n=35 is completely reasonable and common in this type of study.
For things like polling, yes. Bigger is better. But scientific studies deal with the question "does this have an effect." In that aspect, you introduce Type I and Type II errors, and sample sizes play into the chances of making them.
So more samples is worse for a study like this?
Seems counter-intuitive, but I'm not a stats guy.