I once did a complex breakdown of the maths behind Power Attack, but here's the simple (and much-simplified) version.
The standard "benefit" of Power Attack when attacking one-handed is that you take a -1 penalty to hit - or in other words decrease your chance to hit by 5% - in exchange for +1 point of damage, yes? So it's pretty obvious that if that 5% additional miss-chance is equivalent to more than 1 point of damage, you're completely wasting your time with Power Attack.
In other words, once you are doing more than 20 points of damage on average with an attack made one-handed, Power Attack is a really silly thing to do: and that's without taking into account crit confirmations, which can be significant depending on your build.
Doubling the payback for Power Attack when using a weapon two-handed makes it worth doing for longer, what with two-handed fighters doing more damage per blow anyway. Now you get +2 points of damage per 5% decrease in chance to hit, meaning that overall you are doing something worthwhile by using the feat so long as you're not expecting to do more than 40 points of damage in a blow. If two-handed weapon wielders only got 1.5x the amount they deducted from their attack rolls, it would only be worth using up to an average damage of 30 points per blow, rendering it useless at mid levels.
It is, of course, more complicated than that. There will always be a range of AC's that you will hit on anything but a 1 or can't hit on anything but 20, which changes the game somewhat. Dice explosions on a crit also skew the figures, depending on threat ranges.
For the record, I don't really believe that WoTC took all this carefully into account before deciding how Power Attack worked; if they had it would have gone against years and years of demonstrating that they Just Don't Get Statistics... but you never know
If anyone's interested, I made a spreadsheet that allows you to input attack bonuses and damage for a variety of attack routines (up to 4 attack routines with up to 4 iterative attacks each) and compares them in graph output across a range of AC's. It's very useful for quickly answering questions about what your optimal strategy will be when comparing possible attack routines for a prospective build. By all means PM me for a copy
