OK, I admit I have not considered multitries_exponent values that are that large. At an exponent of 15, that’s an extra 1 MB of data that would need to be downloaded; though I suppose if it saves more than 1 MB of total downloading then it’s worth thinking about.
BTW my guess for average windback_periods would be something like 25, not 4. windback_periods = 4 implies that it takes one period to evaluate a collation in one period, which basically means zero safety margin.