You are looking at plea bargains on an individual / Case-by-Case basis.
The problem arises when you view the system dynamics as a whole. The government using plea bargains to minimize the number of cases that go to trial (over 96% of the cases never go to trial). That allows them to spend significantly more on each case that does go to trial. That significant expenditure increase dramatically increases the likelihood of a guilty verdict.
As much as we would like to believe than the amount of money spent by each side doesn't truly determine the outcome of court cases (and elections, patents applications, etc), that would be naive. It's similar to why we need public defenders to ensure a fair trial.
The result is a situation where a minority of cases go to trial, the government virtually always wins them (in large part due to having spent more money on them). That success rate is used to frighten everyone else into taking plea bargains, which reinforces the system.
Just to prove this, imagine if the rates were reversed and 96% of people went to trial. That increased case load would significantly hinder the ability of the prosecutors, and the conviction rate at trial would decrease dramatically. This would then encourage people to go to trial more, causing a reinforcement effect in the other direction.
I'm not necessarily advocating that everyone go to trial and increase the costs of prosecution because the pendulum might swing too much the other way and our justice system could fail to secure important guilty verdicts.
I do think that there is a massive asymmetry currently between public defenders and prosecutors that is to the detriment of society (exploding prison populations, devastated inner city communities, etc). This system of plea bargains is a large part of this asymmetry due to its effects on per case resource expenditure).
This would be an interesting economics / criminal justice model to create. Creating a simplified, theoretical model on the effect of expenditure on verdict, and then using game theory to find the equilibrium plea bargain rates and predicted number of incorrect verdicts (both false positives and false negatives).