My concern was based on there being enough Vbe drop to have Q1 "on enough." What really saves it is the feedback design.
Analysis can be simplified by shorting out R3, in which case the whole thing resolves to a simple feedback design. Being at DC, we can cheat and assume that Beta is "large enough" not to worry too much about, in which case attention can be paid to R1 and Q2.
(Vcc-R1Ic1)=Vt ln (Ic2/Is)
Eventually this falls out to the nice "clean" form of Is*exp[(Vcc-R1Ic1)/0.0258] = Ic2
Let R1 be small, and you have Ic1 having a strong contribution to the total currents into R2. Let R1 be large, and the majority of the currents into R2 are supplied from Q2. The clever part of the circuit is that the feedback drops the required base current needed by Q1, and makes the input look much higher in impedance as an input.
Given the feedback in use, devices are both kept "on" but simply using the 0.7V drops is not going to be sufficient. The answers here are going to rely on the use of the currents and the device gains, of which Ic2 will depend on R1. If R1 is not a component of the total gain, something is wrong.
Edit:
I decided to run this out through lunch.My final equation is:
Vo = (Vin - 0.7) + (Bf/(Bf+1))*( (Vin-0.7)/R2 - 0.7/R1 )*R3
SPICE confirms this for values: R1=23k, R2=1.3k, R3=729 (Test 1) and R1=10k, R2=5.3k and R3=760.3 (Test 2)
The catch is, the "gain" isn't flat for Vin, but rather it is only the value at your "target" Vin set point.
In the above cases, G=1 at Vin=Vo=2 (Test 1) and Vin=Vo=6 (test 2).
For a gain of 2, at 2.0Vin, R1=10k, R2=5.3k, R3=15.559k, Vo=3.983V so the equation checks out.