On second thought, there is a much more reliable and predictable limit to output stage transconductance than storage effect or parasitics: the output protection resistors.
Take the classic NE5532 with its 15Ω. You aren't getting more than 70mA/V out of that thing. Given 8V/µs voltage slew rate, that's 560mA/µs current slew rate into a dead short. So at 5MHz it's limited to 18mA peak into a short, even though normally it could do twice more. Add 15Ω of external load and the limit becomes 9mA at 5MHz or 18mA at 2.5MHz.
But the above example also shows that one needs to go to quite extreme lengths to encounter this problem

Or use RC4558. According to Raytheon, its output resistor is 120Ω

But I'm not sure if it's not a typo and I felt like using such a lousy opamp for this example would be cheating
