As a consequence of making way too many AMVs (importantly, most of them pretty average), I have my own statistically significant sample size to mess around with whenever something like this comes up. I took a snapshot of my current results, then re-sorted them against certain parameters to export and draw on the graphs that follow.

This just tracks hits by video over the SH catalog. This indicates that hits are not simply linear in time, otherwise the curve would be flat down. What this looks like is a recentist bias until the middle of 2004, after which hits go towards linear in time. This may be an effect of what I was working with; videos after the inflection point tend to get a lot more obscure in their sources, and the rate of release slows down significantly.

Just stars versus local downloads. As anticipated, stars track local downloads pretty closely and converge rapidly at lower volumes.

Average star score plotted against total opinions. The nearly flat curve through the approximate averages indicates that the number of opinions a video gets doesn't track with its quality. Note the same heavy-tail behavior seen in numbers of opinions as seen practically everywhere else that it can manifest.

Average star score plotted against total opinions, this time with more unnecessary lines. The low of each opinion-level block is pretty consistent, but the highs go up logarithmically.

The share to which I contribute to the problem. After some laborious math, I worked out that in the .org at large, 48.7% of videos have at least one source on the most-used anime list, and a nondisjunct 18.27% of videos use an artist on the most-used artists list. (These are estimations, the second assuming that each video includes a maximum of one artist on the list.) My comparable figures are 12.71% and 5.8% respectively. The downside of this is that I don't have a large enough sample space to accurately assess to what degree most-used is a contributing rather than just nondisjoint problem. How likely a video with an anime on the most-used list is to have music from a most-used artist, and vice versa, is a problem for another researcher.

I manually flagged each video as having music that was metal or not and underground or not. These are the splits for star rating. Behavior and averages are pretty uniform, but note that the highest highs are with underground material. The underground not-metal low is lower than in the corresponding non-underground split, but the underground metal low is higher than the corresponding point in the non-underground sample.
Yes, underground is arbitrary. For the purposes of this stat, it was decided by the question "Would the average person who both listens to this kind of music and watches AMVs be expected to be familiar with this band's music?". 'No' decisions went into underground, everything else went into the normal bucket.

A messy manual overlay of star score against local download volume with a fit line drawn on it. The weak upward slope is what would be expected if greater volume leads to regression to the mean in scoring. The relationship is pretty weak, though, which indicates that convergence to the mean emerges pretty quickly.

This breaks out number of downloads by genre. Some genres that don't have any representatives on local are not shown in the breakouts. What stands out here is 1) the significant demand for drama 2) the low demand for horror and experimental stuff, and 3) the consistency that the heavy-tailing behaves with across categories with a large enough sample space for it to be visible at all.
Requests for further splits, unless they sound especially interesting, will be met with "make your own 120 data points about it".
--K