Interesting and thought-provoking post. As to what to make of all the little experiments showing a default change here, a reminder there, a nudge there, an incentive there etc. lead to little improvements.
I have two possible responses:
1) We have trouble finding any evidence that any big policy really works that well. So then perhaps development really is a slow process of making small tweak after small tweak to an existing system - and if you make enough changes which each gain you 2-5%, over time they add up.
2) we should think of all these little experiments like venture capital, with every now and then we discover something that delivers exceedingly high returns/big changes (e.g. defaulting to opt-out rather than opt-in on organ donations).
In terms of your point about tax collection, theory might tell us to scrap the entire tax system and start again from scratch (certainly that seems optimal in the US system), but politics makes this impossible - so then small tweaks to an existing system may be the best we can hope for, and then we want to know which ones seem to matter.