Hmm, is that a well-defined property? You can get infinite gain by pushing arbitrarily close to instability (hence the superregen radio)...
That assumes that you're evaluating over all possible input and output coupling impedances.
It may be more practical to speak in terms of maximum stable gain, and gain-bandwidth?
Gain-bandwidth is easily modeled with an ideal transconductance amplifier that has an RC output characteristic, and infinite isolation. The R is due to BJT Early effect (or MOS channel length modulation, or electric field feedback in the vacuum tube, etc.), in parallel with the load resistance, and C is the output capacitance. You can resonate the C with L, but to reach a higher and higher center frequency, you need smaller and smaller L, which drives the Q up (Q = R / sqrt(L/C) ). Thus the bandwidth fraction (in terms of center frequency) drops, but it drops at the same rate that center frequency rises -- the bandwidth remains constant. Bandwidth is inversely proportional to R, so it's also a gain-bandwidth tradeoff.
Maximum stable gain is rather more involved -- I don't know the proof very well myself, but it has to do with solving for the stability criterion over all input and output impedances.
The heart of it is this: if isolation is less than (feedback is more than) maximum device gain, then there will be some condition where the feedback, phase shifted appropriately, will result in oscillation. You then have a conditionally stable amplifier. As you go to ever-higher gains, the stability region, the locus of input and output impedances where it remains stable, continues to shrink.
Tim