The classic instrumentation amplifier circuit consisting of opamps 1 and 2 in your diagram have a common mode rejection ratio of 0db, otherwise called "no rejection at all". In other words, the outputs are still differential, both containing common mode noise, and still requiring a third differential stage to remove that noise. This can be seen here:
simulate this circuit – Schematic created using CircuitLab
V1 produces a 1V low frequency input offset common to both inputs. This represents common-mode input "noise", from a source with some impedance Rnoise. This noise is added to the signal from V2, a smaller and higher frequency 100mV signal, representing the ECG potentials to be measured. It can be clearly seen that the two differential inputs IN1 and IN2 contain this "common mode" noise offset.
Take a look at the outputs:
Here we see than OUT1 and OUT2 (blue and orange) are complements of each other, which both still contain the 1Hz common mode noise. OUT3 is the result of subtracting OUT2 from OUT1. Obviously, what you want is OUT3, a proper single-ended single signal representing the potential difference between IN1 and IN2.
Usually instrumentation amplifier ICs include this final difference stage, to provide a single-ended output, with the common-mode element eliminated.
However, your diagram performs common mode rejection without that third stage, by offsetting the body's potential to half-way between the two outputs. This is bootstrapping the body to have the same potential as the output's mid-point, thereby shifting the potentials of all three participants (body, and opamps) into the same regime. Common mode noise (V1 in my example) is eliminated because the bi-potential amplifier itself is measuring potentials relative to its own, imposed, idea of what the body's potential should be.
Here is that scenario simulated:
simulate this circuit
Without some proper analysis, I can't tell you the significance of different amplitudes of the signals at OUT1 and OUT2, but it is clear that the common-mode noise is gone. And you still benefit from the balanced inputs afforded to you by the instrumentaion amplifier setup.
I don't know offhand what advantage this bootstrapping offers over a simple difference stage following the instrumentaion amplifier stage, but it does work. I imagine that by employing this bootstrapping technique and a difference amplifier for OUT1 and OUT2, you can achieve really good common-mode rejection.
Perhaps there are other benefits that your book can point out. I would be interested to know.