I think this may be just a case of the HP54503A being weird. As you point out, at such slow time/div settings it should be able to fill the entire 500 point record in one pass. There are some odd things in the specifications though:
When doing a simultaneous acquisition on two channels, maybe it can only fill half of the points on each channel on each acquisition. If you trigger on a third source like the line so both channels are asynchronous to the trigger, then this should show up on both channels. I think this is the most likely explanation. (1)
Alternatively the HPIB record length it 500 points in single shot mode and 1000 points in repetitive mode at slow time/div settings indicating that it always take two acquisitions to fill. If the display record is derived from the HPIB record, then what you observed could make sense.
The display record is 500 points but the "data display resolution" is only 451 horizontal x 256 vertical. How does that work? Does the record extend beyond the bounds of the display? How many horizontal points are displayed per division?
As another test, you might try using line triggering so the acquisitions are triggered asynchronously to all input signals.
Apparently it can only simultaneously acquire channels 1 and 4. What happens with other channel combinations? Does it also simultaneously acquire on channels 2 and 3?
(1) The early Tektronix DSOs like the 2230 and 2232 work this way however they also halve the record length for each channel when 2 channels are used so the half length records still get filled in one acquisition. This is not as bad as it sounds because their display with one channel is 100 points/div and even at 50 points/div, they still have higher horizontal resolution than many DSOs now.