One way to achieve a fractional prescale with 1/N granularity is to set up N identical triggers, but all with different trigger IDs, and request a rate for each of them that is 1/N of what we normally request for this trigger. The each copy will get a prescale of N, and the will add up to the required total rate (minus calculable overlaps).

To avoid run-by-run rate fluctuations because of coincidences in the randomized initial offsets, one can request the target rates that form a geometric progression with an irrational common ratio, like pi. :-)