@cpsdqs EDIT:
Maybe "one single jump target at a time (which must be the active instruction when set), with a monotonically-decreasing-without-rollover jump-if-non-zero counter, which can't be forward-jumped over"? (EDIT2: forward-jumps clear the target&counter? no, 'cause we want IFs to be able to conditionally drop the counter /further/... hm.)
Then max time is naively "number of back-jump targets"*"max length before jump"*"max counter of jumps",
but can be more narrowly SUM("length from target to jump"*"counter set at target" OVER "back-jumps")