Yea me and Paul are talking about something with an extra level of complexity, so it's harder to analyze than those examples.

The big difference is when the bouyant 'buckets' get to the top, they don't need energy to reduce their density and make them sink again, in fact we can extract energy while doing this by burning the gas and converting it to liquid water.

I suspect the problem is the energy required to convert this liquid water back into gas at the bottom, but havn't quite seen how that can use so much energy yet.