If I look at ratings I put more weight on those that actually have thoughtful comments. Those with little or no data to support why don't add much value. I do like the comments good/bad about each game to learn about new machines. I'll also consider that the machine they had may not have been setup properly or had electronic or mechanical issues. Also, was the game running the latest code? How were each of the settings configured for a solid state that has settings? Some people initially complain a game is too easy yet some simple tweaks to some settings can make a world of difference!
The way a game is dialed in can help. I initially played some games that weren't setup or adjusted properly and it was a bit frustrating. Later I ended up falling into a deal on the same title that wasn't working, fixed it, and once adjusted it is a great machine.
What I would like to see on the ratings is adding another view by date ranges. The top 100 is a nice guide for overall but can be skewed by people that may just be familiar with newer games and rating older games against those. Maybe some date ranges for like machines (1977-1984) (1985-1991), etc. Ranges can be set to get the most similar groups of games together. Then see the top games in those groups.