"Somewhat poorly implemented"?

Try these two null hypotheses:

• "science" as currently implemented is entirely worthless on the margin (relative to costs of brainpower etc.)
• median "scientist" generates zero net contribution to human knowledge (relative to similar person who is not "scientist"; this claim is far stronger, as it is about median not marginal "scientist")

How confident are you they're not true, and the sign points in the right direction?

My overnight fix idea is impact factor tax - convince people scoring scientists by their publications to automatically slash citation value by some % (50% would be a good start, with clear understanding it will go up to 100% eventually) for every publication not openly available, to break their network effect, and at least make all results open.

That won't get anywhere near fixing science, but I cannot think of anything else with better returns on effort. I haven't heard anyone else proposing this particular idea, so if it turns out to be old it must be by convergence.

I'd estimate that including a lot more serious statistics in education would have more significant effect than destroying pay walls, but I see no realistic way of fixing that right now.