I just had a quick first look at lhe event files produced in multi backend tests.
In dy+3j just a few occasional events are the same, but most are different. The problem is also that the weight is completely different. This is from the 500 events in PR #948
diff events_fortran.lhe events_cppnone.lhe | more
450c450
< # Integrated weight (pb) : 142.38107118268414
---
> # Integrated weight (pb) : 3199.366020218372
455c455
< 1.423811e+02 6.637245e-01 1.423811e+02 0
---
> 3.199366e+03 1.661125e+01 3.199366e+03 0
Will need to understand this.