I'm in the process of taking a closer look at this issue at the request of several others both in and out of government, but I can tell you this much on the basis of what I know from my past research and experience: It would not be terribly difficult to structure a tax, even one that approached a rate of 100%, levied on some or all of the bonuses already handed out (or to be handed out in the future) by AIG and other recipients of federal bailout funds so that the tax would survive bill of attainder clause challenge.------
Such a tax would presumably be leveled on the basis of some criterion sufficiently general to avoid classification as a measure targeting solely a closed class of identified and named individuals. The fact that the individuals subject to the tax in its retroactive application would in principle be readily identifiable would not suffice to doom the tax either from a bill of attainder perspective or from a due process perspective. Moreover, the fact that the aim of such a tax would be manifestly regulatory and fiscal rather than punitive and condemnatory, and that the tax would be part of a measure that would be prospective as well as retroactive in its operation, would serve to blunt the force of any bill of attainder challenge. Finally, such a tax would be devoid of the sting of political retribution and would not partake of the classic "trial by legislature" that the attainder ban was designed to avoid.
All things considered, I believe it very likely that Congress could design a fully constitutional means of clawing back into the federal treasury all amounts paid (or to be paid in the future) in the form of retention bonuses from federal funds disbursed either by the Federal Reserve Board pursuant to legislative authorization tracing to the 1930s or by the Treasury pursuant to the most recently enacted federal bailout and stimulus measures.
Further, for an interesting analysis on whether the government can retroactively invalidate the AIG contracts which contained the bonus provisions for executives, take a look at this article by Lawrence Cunningham.