Hold on, you expect 250-500ns control pulses to generate ~ 1000ns light pulses? Have you established this latency through measurements?
I'm not sure if you got the gist of my comment, but I foresee problems with 250-500ns pulses, and certainly with pulses shorter than 100ns since they are either close to, or even below, the switching speed of the led driver, which apparently is <100ns / 10 MHz. There may be problems with e.g. overshoot and ringing that result in the light output scaling extremely non-linearly with the pulse period.
See also further on page 7:
The LEDs and the CCR switch extremely fast, less than 100 nanoseconds.
It's not entirely clear if this implies that the control frequency is actually 10MHz, or that the switching propagation delay is 100ns, which would imply a far lower control frequency (let's say 1MHz and lower). In my experience, integrated led current drivers tend to work at frequencies of 100kHz - 1MHz. Trying to use pulses under 1us seems rather problematic to me if any kind of consistency is a requirement.
Yeah, but don't 
Been there, done that. Besides, I doubt it'll be fast enough for your application, regardless of space requirements etc.
That would help, indeed. Frankly, I'd investigate that route a bit more instead of the led drivers you currently use. I don't think they're quite up to the job you have in mind for them. Not in the future project, but also not in the current 'slow' project that apparently requires 250-500ns pulses. In your place I'd brew a nice cup of tea and spend an afternoon playing around with Spice.