Good question — I have settled on trading a little raw speed for isolation and a clean transient, and that has worked better than chasing the fastest datasheet number. For sub-6 GHz antenna diversity I usually use an absorptive SOI SPDT with about 25–30 dB isolation, under 0.5 dB insertion loss, and a switching time of roughly 1–2 µs including driver and settling, rather than a 100 ns reflective switch that rings.
Isolation matters most on the receive side. A leaked transmit signal through a 15 dB switch can desensitise the idle path and, worse, re-radiate through the other antenna, so the diversity gain you hoped for disappears. I measure switching time from the control edge to the RF envelope settled within 0.5 dB, not to the datasheet 50% point — that difference alone has been 300–500 ns on my bench.
Two things that broke for me: first, a reflective off-port presented a near-open to the LNA input and created a standing-wave ripple that moved with cable flex; an absorptive switch fixed it. Second, switching during a TDD slot without blanking the PA produced a spike that the receiver saw as a burst error. Now I blank the PA, wait for the switch to settle, and only then re-enable — the blanking interval is set longer than the worst-case settling time over temperature.
If your diversity is TDD-aligned, I would start with isolation and blanking, then check whether 1 µs is really too slow for your slot timing. What band and slot length are you working with?