Normally, the order subsystems start doesn't matter a lot-- except if memory is tight.
If your subsystem descriptions in total will claim more memory than physically available your subsystem will start with whatever memory is available.
If no more memory is available, I don't recall if the subsystem won't start or if it gets dumped in *BASE.
We use Robot's AutoTune utility, which lets us set a lower and upper limit for each subsystem. The sum of the lower limits will fit in available memory. AutoTune handles the battle of the subsystems to see who gets the remaining memory.
We don't use IBM's tuner, so I don't know if the same lower/upper limit trick will work there or not.
One other reason the order subsystems start could matter is if more than one subsystem tries to allocate the same devices (ie interactive sessions). We have several interactive subsystems, with one dedicated to the 'laser tag' hand-held scanners used on the shop floor. Since the normal QINTER* subsystems allocate all interactive devices, we start the laser-tag subsystem late in the IPL process, so it can 'steal' the devices back from the QINTER* subsystems. [And, no, we haven't looked at the names or device types recently to see if we could not allocate those devices in the QINTER* subsystems!]
Paul E Musselman