Put another way: it's essentially a phase-shift oscillator (notice the three RCs in the feedback loop), so has extraordinary gain at resonance, which can be far above the asymptotic "skirts", nothing wrong about that. What is... not necessarily wrong, but tricky anyway, is the loop gain must be very slightly less than unity for this to happen; more and it simply oscillates as usual, less and the gain peak and therefore selectivity is far less exaggerated (but, that may have sound-shaping relevance too, assuming the change in overall gain itself is compensated for; that is, this is a variable one can tune to adjust the sound quality for various purposes).
Gain depends on exact resistor values, and transistor bias, so it's tricky to balance.
This is also the basis of the regenerative receiver; but not however the superregen, which harnesses an added layer/quirk on top of it.
Tim