Atrios' answer to Paul Waldman's question is correct, but there is actually a meaningful economic point to being more concerned about marginal rates than just total tax bill. The reasoning is incentives, and the idea is sound enough, but it is just really stupid to apply it to the US and our current tax code.
Basically, if someone pulling in $500k/year and paying $130k in taxes works a bit harder, and makes an extra $50k, then the extra taxes are (essentially) that times their marginal rate and that's it. If you adjust things so that their taxes go up to $200k, but their marginal rate remains exactly the same, then their incentive to work harder and make more is unchanged. If, on the other hand, you keep the loopholes, but increase the marginal rates to offset the same difference, now, the incentive to working and earning more has been reduced (possibly by a lot).
The problem with this argument in the US is that the incentives are already very backwards. As much as the federal income tax is progressive, most state taxes are flat to regressive, same with city, and wage taxes are--by virtue of SS not applying to anything over $106k--regressive. On top of that many social insurance programs provide diminishing benefits as income increases, which sets up some points where the effective marginal tax rate exceeds 100%, and that is all at lower incomes. The result of all this is that marginal effective rate is pretty flat (~40%) for everyone except the very poor (where effective marginal rates are near 100%) and the very wealthy (where it is 15%).
An extreme show of the negative incentives at low income is in the chart below, where a single mom with $29k in income has to more than double that to regain the same standard of living.
I would like to note that the presentation that comes from (PA sec of public welfare) is, despite this very nice graph, not really good. The problem with our tax and transfer programs is that they are very complex, and feature abrupt ends to many benefits. But this is the case because the better (simple, gradually diminishing benefits that leaves no cliff) solution is not politically viable.
A simple cash transfer program through the tax code would be great, but people have strong objections to giving no-strings-attached money to [poor] people. So we have to give food stamps (that can't be used to buy clothing/housing) and then a housing program (that doesn't help with health care) and then a special insurance program... On top of that they have hard cutoffs, because "clearly someone making $x per year doesn't need help with [y]" but since non-heartless, non-assholes want to provide as much help as possible, that assistance is provided right up to the cutoff (also, with something like medicaid, it is really hard to taper it off without making it insanely complex...and without being a total prick).