I have a problem with the latency between some parts of my patches - some of them are delayed by 64 samples (=block size). I think I've created some [r~]s before other [s~]s, which would lead to this behaviour. But: I don't know which objects are affected and I've got more than 50 [s~]s and [r~]s ;-)
see /doc/3.audio.examples/G05.execution.order.pd
it depends on the way the dsp chain is sorted ... the solution is, well, not really clean imo... (expecially, because it can add a computation overhead for an unneeded connection)
cheers ... tim