The isolation-bandwidth trade-off is real, and in practice I have never seen a single passive circulator carry a full-duplex link on its own. A ferrite circulator is usually good for 20-25 dB of isolation over a few tens of MHz, with insertion loss around 0.3-0.5 dB, and the isolation falls off quickly at the band edges because the ferrite resonance and the matching network both move with frequency. If you need 40 dB or more across a wide band, plan the budget in layers rather than asking the circulator for all of it.
The layering that worked for us was: antenna isolation first, then the circulator, then analogue cancellation, then digital self-interference cancellation. Each stage only needs to be good at what it is good at. The circulator handles the bulk of the direct leakage, a simple analogue canceller with a tapped TX sample removes the dominant leakage path including the antenna reflection, and the digital stage cleans up the residual and the nonlinear products. Trying to squeeze another 10 dB out of the circulator usually cost more in insertion loss and size than adding a modest analogue canceller.
Two things to measure before choosing a part: first, isolation versus TX power and temperature, not just small-signal S-parameters, because ferrite isolation drifts with both. Second, isolation with your actual antenna attached, not a 50 ohm load, because the antenna return loss feeds directly back into the leakage path. What bandwidth and TX power are you targeting, and how much isolation can your digital cancellation already provide?