OPA2891: Crosstalk, asymmetry between channels (datasheet figure 5-21)

Tool/software:

I am considering using OPA2891 for a high-speed servo (dc to100 MHz, but with the OPAx891 being mostly responsible for the signal component at or above about ~10 MHz). And, this high-speed analogue domain being the one where I have most uncertainty of meeting the design requirements for my servo.

I am pondering between use of OPA2891 (to reduce both board size and cost) and OPA891 (to have simpler ability to fine-tune the dc offset build-up in my fast signal path before branching off to slower precision op amps for the content below 10 MHz and 1 MHz, respectively; and to not have to worry as much about any inside package cross-talk of signal from my separate rf input path to the low-noise signal pathway). 

The solution with only OPA2891 is very tempting, but the datasheet crosstalk figure for OPA2891 is only up to 10 MHz, and it further shows that "Ch-A to Ch-B" has much less crosstalk than vice versa from "Ch-B to Ch-A".

1. Is this massive difference real, or just an effect of a very small sample?

2. Which direction is this? Is Ch-A Ch-1, as in a large signal going through Ch-1 causes much less error signal in Ch-2 than a large signal in Ch-2 causes in Ch-1?

3. Do you have the data for the typical figure past 10 MHz?

Any reply, even to just one of the three questions would be greatly appreciated as it would simplify my constraints on which two signal parts can share a physical package, and would inform me if I should use Ch-1 or Ch-2 for the more critical signal side at each point in the circuit where cross-talk can be an issue.