The question of average speed in the Paris-Roubaix is interesting, not least because the route is similar each year and it's pancake flat - so wind should matter. I'm a little late to the conversation because I wanted to find the relevant data, which was not simple. I took the trouble of downloading the French meteorological data from the Paris going back to the 1850s. I also pulled the Paris-Roubaix data (distance, time, average speed, date). I picked up the wind speed and wind direction data (and temperature) for the race dates. I couldn't find rainfall data which might have been helpful too. Data for some years were missing but it gives me 102 race editions with wind speed data. I classified wind as tail wind if it was between South-West and South-East, head wind if it was between North-West and North-East, and crosswind otherwise. I did a simple enough regression analysis and the input data did a reasonable job of predicting the observed average speeds.
I made some nice plots but it seems you can only share images if they are already online, so you'll just have to trust me on this.
What I found was that race distance, time (i.e., year), and the combination of wind direction and strength matter. Shorter more recent editions with a strong tailwind are fastest. At a typical windspeed of 4m/s, a race with a tailwind will be 5km/h faster than with a headwind. Under the same conditions, the race today runs about 3km/h faster than in 2000. So based on what is happening generally, the average speed this year was not out of step with what you'd expect given the conditions.
Interesting to note that the editions with the most extreme windspeeds were generally quite a long time ago. The most recent edition with strong winds was 1990 with a 7m/s headwind and the average speed was only 34.9km/h. It might explain the photo finish on that one.
The most interesting thing to me is that the trend in average speed was largely static pre-WW1, increasingly steeply from then to WW2, slightly downward from 1943 to 1990, and increasing steadily since then. I'm sure the timing with the advent of EPO is entirely coincidental. Of course, it also links in with increasing use of carbon and aero designs and the modern habits of bringing your own ketone-impregnated mattress to use on the post-race turbo trainer, so who knows?