They should yield (approximately) the same result. But, to be pedantic, you haven't accounted for latency within the network.
Somebody should be whipped, either for:
2) You, for making even this aged arch-pedant wince. :-)
Ding!
Seriously, can I also add that RADIUS interim accounting is almost essential in this scenario. Real world accounting and session boundaries mis-match badly making it almost mandatory to use interim accounting records to get an approximation of what the figures look like from a billing perspective. I'll also add "watch out for missing records" - I've found RADIUS to be the lossiest network protocol per foot of cabling that I've ever used.
I can't say I've seen this. Having collected hundreds of millions of radius packets in my years (hell, we were running PM-2e's in 1996), and have written several accounting collectors, I can't say I agree. If you follow the specifications properly, unless you have issues with the transmitting device (read: BUG), RADIUS accounting has always been good to me. And, I've not seen the behavior you describe that requires interim.