Bail tends to be set in relation to the criminal record of an individual, the seriousness of an offense, how much of a flight risk they are, and whether or not they're likely to be a danger to themselves or others if released. It's not the 'norm'. So for instance of all individuals (arrested for any offense) 77% of those with no priors end up being released without financial condition. For those with misdemeanor priors it's 63% and 46% for those with felony priors.
And there are excellent test cases that can experimentally show whether these release conditions are reasonable. For instance there are occasional emergency releases caused by things such as prison overcrowding. And this is where people that would not normally be released without condition, are. Of these individuals 52% end up being charged with pretrial misconduct! For the general population of people intentionally released that figure ranges from 27%-36% depending on the release type.
And when somebody is forced to pay bail, they have numerous options. Generally they only have to pay 10% of the cost of the bail. In some cases that can be paid directly to the court who will then refund that 10% once they return for and complete their trial. If that's not an option the way bail bondsmen work is that an individual pays the bail bondsmen 10%. The bail bondsmen then go to the court and pay 100% of the bail on behalf of the individual. They get that 100% back once the individual returns and completes their trial, so the 10% is their profit margin. This is also where modern day 'bounting hunting' tends to come into play -- the bail bondsmen have a very strong incentive to ensure that their clients do not flee.
---
So the point of this all is that in the vast majority of cases we do simply let people go on the assumption they'll return to trial. For those that we don't let go, there tends to be a pretty good reason. That article from the BJS I linked has a great table at the bottom that offers some regression figures for release/return rates of various groups helping to determine how justified various decisions are. In most cases they are, in some cases they are not. It's pretty reasonable to conclude that it's a system that does well, but could be made to do better. But it's quite different from what you would portray it as.
I'm trying to find a more comprehensive article containing data across both misdemeanors/felonies and state/federal courts, but running up short. Their data seems to be heavily segmented in not so useful ways. If you find something along these lines I'd very much appreciate if you'd ping it regardless of its direct relevance.
One other thing you have to look at as well is the time frame. As time goes on you get closer to the real 'risk' rates. That's also detailed at the top of page 7. For instance only 8% of those those that would be rearrested were rearrested within a week, which goes to 29% in a month, 62% in 3 months, and 85% in 6 months.
Again it's very disappointing that the data aren't more clearly quantified and classified across documents.
A basic principle of common law and the American justice system is that it's better to free the guilty than imprison the innocent. So just on American principles, letting poor people go on the assumption that they'll return for trial is better than keeping them in jail for lack of bail.