The speed of the serve is measured at the point of initial contact. If it were measured when it crossed the net, you'd rarely have a serve exceed 100 MPH. And the serve loses fuly half of it's initial speed by the time it reaches the opposing baseline. (Slightly more until it reaches the returner.) That's why the readings on serves and groundstrokes vary so widely. The most commonly used groundstroke measuring devices clock the average strokes of their measured group (almost exclusively top 50 ATP pros, with emphasis on top 10) in the 45-50 MPH range. It doesn't look very impressive when compared to a "140 MPH" serve, but if you consider that said serve is traveling less than 70 MPH when it reaches the returner, it lines up. (Much like "140 MPH" badminton shots are only actually traveling ~40 MPH when they cross the net.)
Getting back to why the majority of the faster serves might be registered on hard courts, the characteristics of the surface ome into play. The lower friction of grass rewards slice. The higher bounce of clay rewards topspin. Thus, those particular serves would be very effective on those surfaces. And spun serves read slower than flat ones.
The heat consideration is an interesting one. In addition to the air temperature, hard courts yield reflected heat as well. This could add a bit more zip. (And also possibly have an effect on radar guns.)
Looking at the current player guide, it lists the top 46 fastest servers from 2001. the list breaks down this way: 13 on grass, 4 on clay, 29 on hardcourts. You'll notice there are no indoor events listed. This is primarily because the list runs through September 16th, and thus skips the fall indoor season.
Out of the rest, the vast majority of the readings are on hardcourts. However, hardcourts comprise ~70% of all tournaments, particularly at the Tier II+ level. (None of the listed serves is from below a Tier II.) There are only 4 grass court events total, and 2 T2+, making grass less than 1/15th of total tournaments. Yet it accounts for more than 1/4th of the fastest readings. Going further, 4 of the top 5, and 7 of the top 9 were on grass.
If anyone has a similar list from the ATP, I'd be curious how theirs broke down by surfaces.
Any physicists around to further the discussion?