An ex-girlfriend of mine had an uncle who worked in the back offices at Caesars Sportsbook and we had a conversation about this once. Here is what he told me...
1. The idea that the media knows exactly how much is being bet on one before the game is a pretty far off. Books will often release the numbers beforehand as a way to encourage more action one way or the other. And those numbers aren't totally accurate. For instance, if you hear "80% is being bet on NE" on Friday, quite often it's 76% and that was just the action up until Wednesday...the money have have actually moved a little the other way at some point and the public isn't totally aware yet.
2. Even if 80% of the money is going to NE at -3, that's not the whole story. People have been making different wagers all year that will be determined by this game. There is often big money placed future bets, such as "The Broncos to win the SB (or AFC Championship) at 8 to 1". Those payouts might be much less compared to "The Patriots to win the SB at 3 to 1" because a huge number of people took NE to win it all back in the preseason. And we don't know how many bets were on the money line either, which has much different odds. His point was there is more to the payouts than just the winner and the spread. It's not a huge factor but casinos know it and keep it in mind when determining their risk.
3. Sportsbooks can get burned by shifting the line. For example, $10M gets bet on NE at -3 on Monday and the books feel its too much action that way (especially if it's the sharps money), so they move it to -4 on Tuesday...money keeps coming in on NE so they push it to -4.5. That forces a lot of action on Denver. That's good, right? Not if the Pats win by 4. That is their nightmare scenario and it happened in SB XIII. The line opened at Pitt -3.5 and all the early money went there. It moved to -4 for a short time and then to -4.5 by kickoff. Nearly all the late money came in on Dallas. The Steelers won 35-31 and it's still referred to as "Black Sunday" in Vegas.
4. Lastly, as many argue, sportsbooks do like to have a 50/50 split because the can't lose. But sometimes they actually don't mind "rolling the dice" like a normal fan. As a few posters pointed out, the casinos are making a fortune either way that day from alcohol/food sales, normal table betting, etc. They love the total volume of the action. And when the only big betting games that day are the SB or two conference games, it's almost impossible to make the numbers 50/50 because there are only a limited number of point spreads to shift around. In the end, they look at the risk as being part of the business and may feel Denver at -3 is truly the most accurate spread. And remember, even if NE covers, they're still getting the vig from the losing money and like I stated in numbers 1 and 2, the number they actually pay out may not be nearly as high as reported after calculating all the betting scenarios.
This is what he told me and it makes sense. But he was kind of a psychopath too, so....
Edit - Spelling