The real strange thing is why they didn't compare their new capacitive coupling with a classic wired connection between the PA and the antenna. Instead they introduce an additional PA with corresponding power consumption when they test the wired connection.

The difference between the standalone chip antenna, with a maximum size of 1 mm, with the proposed antenna, with an size of 17 mm, is not revolutionary, it is expected due to the very bad efficiency of electrically small antennas.

Yes, it is counterintuitive. And also not what is actually claimed in the paper.
In the paper three designs are compared:

(1) One with only an antenna on chip. That is, an antenna on the actual chip, with a size of 1x0.5 mm. Draws 3.3 mW, "range" 1m.
("Range" is a very strange measure in RF design...) (2) The same chip but without the on-chip antenna. Instead the power is coupled to an additional PA-amplifier, and an external small folded dipole antenna: Size about 16x10 mm. Draws 38 mW, "Range" 75 m. (3) The same chip withou the PA, with the on-chip antenna coupling to an external patch antenna of size 17x17 mm. Draws 3.3 mW, "Range" 24 m.

In summary: Nice engineering work, but no conclusions can be drawn, as it is very much a case of apples and oranges. (No constant TX power, No constant size, Not very much constant between the designs at all.)

And a classic mobile phone does not use an on-chip antenna at all. So this design will not give any benefit to your iPhone or Blackberry etc.

You can not get any gain in an on-chip antenna at this frequencies: it is to small. He is comparing the use of only an on-chip antenna, which is never used in mobile phones, with the use of a coupled external, somewhat bigger, antenna on a ceramic substrate. Not at all suprising that he gets a better performance with the latter, as it is bigger. He would get even better performance with a classic mobile phone antenna, though.

I.e. This will not revolutionize the battery life of your iPhone or Blackberry. The losses in the coupling between the integrated PA and the antenna are very small (if we disregard detuning due to human proximity effects. Which is another story, and which is not influenced at all by the design in question.)

The comparison between two different antennas at different powers is not very good science - it is somewhat suprising it got published. (But it is only at a small conference, so it is not that surprising.)

Minor nitpick: 2.4 GHz is not the frequency of maximum aborbance of water. The frequency of the maximum is temperature dependent, and the absorbance peak is very broad. Thus there is no need to use any special frequency. 2.4 GHz is used in microwave ovens due to that it was free to use, being an ISM band, and that the penetration depth is useful for cooking.