The issue of whether Ricardian equivalence is a good approximation is closely connected with the issue of whether the permanent-income hypothesis provides a good description of consumption behavior. In the permanent income model, only a household's lifetime budget constraint affects its behavior; the time path of its after-tax income does not matter. A bond issue today repaid by future taxes affects the path of after-tax income without changing the lifetime budget constraint. Thus if the permanent-income hypothesis describes consumption behavior well, Ricardian equivalence is likely to be a good approximation. But significant departures from the permanent-income hypothesis can lead to significant departures from Ricardian equivalence.
We saw in Chapter 7 that the permanent-income hypothesis fails in important way's: most households have little wealth, and predictable changes in after-tax income lead to predictable changes in consumption. This suggests that Ricardian equivalence may fail in a quantitatively important way as well: if current disposable income has a significant impact on consumption for a given lifetime budget constraint, a tax cut accompanied by an offsetting future tax increase is likely to have a significant impact on consumption.
This discussion suggests that there is little reason to expect Ricardian equivalence to provide a good first approximation in practice. The Ricardian equivalence result rests on the permanent-income hypothesis, and the permanent-income hypothesis fails in quantitatively important ways.
There's something to be said for the assumptions of Ricardian Equivalence. You don't need people to be infinitely rational, only that deviations happen on both sides with equal probability. As much of the behavioral literature focuses only on the existence of deviations (Gambler's fallacy causes us to bet counter to the trend, while hot hand has us follow momentum), we don't have a great sense of whether distortions are systematically biased in equilibrium. The sort of crazy assumptions required could very well be reasonable approximations, ones that are consistent with the degree of foresight people seem to have--people seem to save for the future, leave bequests, and have cut down smoking and pollution drastically in response to more information.
Still, the empirical evidence is pretty lacking here. As someone once said, Economics is not Mathematics or Music; models not pretty or useful by themselves, but are only valuable to the extent they accurately represent reality.