analogRead() seems to be fast enough. ... But we're drifting off topic...
Well, not really, because timing seems to be important to you.
-My sample clock wants to be 125us which I think will be another timer interrupt (I'll write that part tomorrow).
A normal analogRead (save fidding with registers) will take 104 uS. So most of your 125 uS is gone in doing the analogRead alone. I'm drawing that to your attention in case you think it is instant, and that you have something like 110 uS over between reads.
One possible thing you could do which would remove the worry about interrupt latency, and also whether or not another interrupt was active, would be to use the external ADC trigger (page 252 and 266 of the datasheet). You could start the conversion highly reliably on a timer compare match event, and then just use an interrupt to gather the ADC results, when available. This also means you don't wait 104 uS doing nothing. Win win!