I’m not sure why it would be multiples of four, but I feel that there is probably an overhead in having each frame as a new repeat of the routine. I’ve just copied my own Change Blindness RSVP experiment from December last year to my demo namespace so you can take a look at the code. The entry in my online demos post points back here for the discussion thread. I’d be interested to see if its timing is good enough, if you’re happy to test it.
Change Blindness RSVP code | try it | online demos
An experiment created in December 2020 for Sasha Andrade (supervised by Michael Pilling). In this case the polygons for each RSVP stream are set up in one routine and then displayed in a second using AutoDraw manipulations to show and hide each frame.