I've nothing very definite in mind, but such inconsistency, if not due to genetic heterogeneity in the inoculum, suggests the possibility that some heterogeneity is arising during the PO4/As-limited growth period, some cultures stochastically acquiring the ability to pass the barriers by some standard or non-standard (mutational) process. Perhaps you need a way to cell-sort that could separate fast-growing from slow-growing cells in a culture that is in the process of escaping the growth limitation.I'm pretty sure the discordant results aren't due to mutation, for two reasons.
First, the #4 cultures that grew quickly appeared to do so with no lag and no period of slow growth during which mutations could arise. The #4 cultures became turbid just as fast as the control #3 cultures without arsenic, so I think that all of the cells in the inoculum contributed to the growth.
Second, one of the experiments where #4 didn't grow (expt. 2) was begun with cells taken from the fast-growing #4 culture of the previous experiment (expt. 1), rather than from the frozen stock of phosphorus-depleted cells used for the other three experiments. If the growth in expt. 1 was due to a mutant clone, the #4 culture in expt. 2 should also have grown fast, but it didn't.