Hello Sol,
I do appreciate your answer, and I am sorry I am late reacting to it.
It looks like we have to cope with the capacitor solution because, as I see it, your other suggestions is not possible to implement because perhaps I have not given you all parameters.
Our system is a single master system and that takes away a few of your ideas. Since we cannot control the clock-stretching of the slave peripherals, we just have to make sure that the capacitors takes care of the colliding signal flanks even if your idea has merit.
Let me tell you what we did to solve this.
Since we have a lot of pins free on the LPC1769 we actually split the I2C system in two. One part "locally" and one part that talks over the opto-couplers.
On the "opto-branch" we put some capacitors on the data lines (yes one send and one receive line due to a I2C driver on each side of the opto-couplers), and changed the duty-cycle of the SCL to 33/66 % to give the slaves extra time to send a stable data.
This actually increased the immunity by a factor of four! and put us clear of the EMC demands.
Again Sol, thank you.
Regards
Mike