To conduct the study, researchers placed 40 computers onto Tor's network as relay machines (the anonymizing service relies on volunteer's computers moving encrypted traffic around). The team then used web-crawling software to categorize the (approximately) 45,000 services that were using Tor's hidden services (which is distinct from the Tor anonymizing network that stops websites from tracking your identity) at the time. Data from their relays provided the insight into where traffic was going, after discounting automated "botnet" activity (usually relating to malware, viruses and so on). This data was presented at the Chaos Computer Congress in Hamburg, Germany, and you can watch the presentation itself here or embedded above.
To further illustrate how out of whack the numbers are, it's estimated that the number of sites -- like Silkroad -- selling contraband, constitute about a quarter of all hidden services. Child abuse sites? Just two percent. This imbalance could be tough love for those who defend services like Tor for its legitimate uses (privacy and security). Though there are unsurprisingly suggestions that the data is open to interpretation, and that web use/behavior patterns need to be factored in (daily visits over monthly purchases etc.). Also, it's possible that the traffic to such sites might not always result from an actual "consumer." Law enforcement will be investigating such sites, and there's the possibility some of the traffic is a noble attempt to knock them offline through denial of service (though it's not clear if this would have been ignored by the study's botnet profiles). There's also an argument that some services are short lived on the deep web, so the study may just be revealing that abusive sites are more persistent.